arX

iv:2

002.

0349

1v2

[cs

.IT

] 3

Aug

202

01

Massive Access for 5G and BeyondXiaoming Chen, Senior Member, IEEE, Derrick Wing Kwan Ng, Senior Member, IEEE, Wei Yu, Fellow, IEEE,

Erik G. Larsson, Fellow, IEEE, Naofal Al-Dhahir, Fellow, IEEE, and Robert Schober, Fellow, IEEE

Abstract—Massive access, also known as massive connectivityor massive machine-type communication (mMTC), is one ofthe main use cases of the fifth-generation (5G) and beyond5G (B5G) wireless networks. A typical application of massiveaccess is the cellular Internet of Things (IoT). Different fromconventional human-type communication, massive access aimsat realizing efficient and reliable communications for a massivenumber of IoT devices. Hence, the main characteristics ofmassive access include low power, massive connectivity, and broadcoverage, which require new concepts, theories, and paradigmsfor the design of next-generation cellular networks. This paperpresents a comprehensive survey of massive access design forB5G wireless networks. Specifically, we provide a detailed reviewof massive access from the perspectives of theory, protocols,techniques, coverage, energy, and security. Furthermore, severalfuture research directions and challenges are identified.

Index Terms—B5G, massive access, cellular IoT, low power,massive connectivity, broad coverage.

I. INTRODUCTION

The widespread applications of the Internet of Things (IoT)

in a variety of fields, e.g. industry, agriculture, medicine, and

traffic, have spurred an explosive growth in the number of IoT

devices [1]-[4]. As of 2017, there were 8.4 billion connected

devices across the world. It has been predicted that this number

will surpass 75.4 billion by 2025 [5], [6]. This growth rate is

tremendous and will further increase over the next decade. It is

also believed that the number of IoT devices will eventually

reach hundreds of billions with a connection density of 10

million devices per km2 by 2030. This trend acts as the

catalyst for speeding up the evolution of IoT to the Internet-

of-Everything (IoE).

To allow IoT devices to connect, interact, and exchange

data anywhere and anytime, they have to be interconnected

wirelessly [7]. Hence, wireless access technology providing

reliable communications is the key to unleash the potential of

the massive IoT. Currently, IoT devices access various wireless

X. Chen is with the College of Information Science and Elec-tronic Engineering, Zhejiang University, Hangzhou 310027, China (e-mail:chen_xiaoming@zju.edu.cn).

D. W. K. Ng is with the School of Electrical Engineering and Telecommu-nications, the University of New South Wales, NSW 2052, Australia (e-mail:w.k.ng@unsw.edu.au).

W. Yu is with the Department of Electrical and Computer En-gineering, University of Toronto, Toronto M5S3G4, Canada (e-mail:weiyu@comm.utoronto.ca).

E. G. Larsson is with the Linkoping University, Dept. ofElectrical Engineering (ISY), 58183 Linkoping, Sweden (e-mail:erik.g.larsson@liu.se).

N. Al-Dhahir is with the Department of Electrical and Computer Engi-neering, the University of Texas at Dallas, TX 75083-0688, USA (e-mail:aldhahir@utdallas.edu).

R. Schober is with the Institute for Digital Communications, Friedrich-Alexander-University Erlangen-Nurnberg, 91058 Erlangen, Germany (e-mail:robert.schober@fau.de).

TABLE ICOMPARISON OF EXISTING WIRELESS SYSTEMS SUPPORTING IOT [15].

Zigbee Bluetooth WiFi LoRa Cellular

Spectrum Unlicensed Unlicensed Unlicensed Unlicensed Licensed

Connectivity Moderate Small Large Massive Massive

Throughput Moderate Low High High High

Range Short Short Moderate Long Long

Security Moderate Low Moderate High High

Power Low Low High Low Low

Mobility No No No Yes Yes

Latency Low Low Low Low Low

networks mainly via low-cost commercial technologies such

as Zigbee [8], Bluetooth [9], and WiFi [10]. However, these

technologies only support short-range wireless access for a

moderate number of devices, e.g. a few hundred devices in

an indoor environment or in a small area. Newly emerging

services require IoT devices to have seamless access over

a wider range. Thus, the existing technologies can only be

adopted as an intermediate solution for serving a small number

of IoT devices, but eventually will become a bottleneck

for providing reliable wireless access to a massive number

of IoT devices. On the other hand, a ubiquitous wireless

infrastructure is a key enabler for realizing wide coverage

for the IoT. Currently, long range radio (LoRa) and cellular

IoT are two main access technologies for low power wide

area networks (LPWAN) [11]-[13]. Compared to the LoRa

technology, cellular IoT is more beneficial and economical for

service providers as it reuses existing cellular infrastructure.

In order to support massive access with a connection density

of 1 million devices per km2 with cellular networks, the 3rd

generation partnership project (3GPP) has selected massive

machine-type communications (mMTC) as one of three main

use cases of 5G wireless networks and provided a dedicated

specification for cellular IoT in Release 13 in 2015 [14]. In this

specification, cellular IoT is categorized as narrow-band IoT

(NB-IoT) for fixed and low-rate scenarios and LTE-machine

(LTE-M) for mobile and high-rate scenarios [15]. Hence, the

existing cellular network architecture and technology can serve

as a solid foundation for enabling massive IoT in practice.

A comparison of existing wireless systems supporting IoT is

provided in Table I.

The key to supporting massive IoT in cellular networks

lies in designing appropriate multiple access techniques. In

fact, enabling multiple access with limited system resources

is an inherent issue in cellular networks. Previous and current

cellular networks have employed a variety of effective multiple

access techniques, such as frequency division multiple access

(FDMA) in the first-generation (1G) wireless networks, time

division multiple access (TDMA) in 2G, code division multi-

ple access (CDMA) in 3G, and orthogonal frequency division

2

multiple access (OFDMA) in 4G and 5G [16]. However, it is

not a trivial task to realize massive access in B5G wireless

networks. First of all, there is a lack of information theoretic

concepts for the design of massive access. In particular, con-

ventional information theory commonly focuses on multiple

access scenarios with only a small number of devices [17],

[18]. It is not straightforward to extend the conventional

multiple access theory to massive access. In particular, short

packets are usually employed in massive access for reducing

access latency and decoding complexity at the receivers, which

requires a much more sophisticated multiple access theory

[19]. Secondly, the commonly adopted grant-based random

access protocols may lead to exceedingly long scheduling

delays and large signaling overheads [20], [21]. In fact, for

grant-based random access protocols, each device would have

to choose a preamble from a pool of orthogonal sequences for

accessing the wireless network. Due to the limited coherence

time and sequence length, the number of orthogonal sequences

is finite. As a result, in the context of massive IoT, two

or more devices would choose the same sequence with a

high probability, leading to collisions and failure of wireless

access. More importantly, the access delay inevitably increases

as the number of devices increases. Thirdly, most existing

IoT networks adopt orthogonal multiple access (OMA) tech-

niques [22]. For example, NB-IoT employs single carrier

frequency division multiple access (SC-FDMA) for the uplink

and OFDMA for the downlink. Although OMA simplifies

the transceiver design, it leads to a low spectral efficiency

in general [23]. In the context of massive IoT, applying OMA

over limited radio spectrum is challenging due to the resulting

underutilization of the system resources. Fourthly, coverage is

a critical issue for low power IoT devices. In order to prolong

the battery life of IoT devices, their transmit powers are

usually very small, e.g. 23 dBm for NB-IoT [24]. As a result,

the received signal is generally weak for signal detection if the

distance between the base transceiver station (BTS) and the

device is large. As a remedy, NB-IoT enhances the coverage

by adopting re-transmission (i.e., time diversity) and low-

order modulation (BPSK/QPSK). These techniques enhance

the service coverage at the cost of inefficient utilization of the

system resources. In other words, if the number of IoT devices

is large, the limited system resources may be insufficient for

wide coverage. Fifthly, security issues of massive access have

to be investigated carefully. Due to the broadcast nature of

wireless channels, confidential wireless signals may also be

received by unintended devices, resulting in potential infor-

mation leakage [25], [26]. Traditionally, cryptography-based

encryption techniques are employed to guarantee security

for wireless access. However, due to the fast evolution of

eavesdropping techniques in recent years, providing secure

encryption has become much more challenging. Unfortunately,

most IoT devices have limited computational capability, such

that they cannot utilize sophisticated encryption techniques.

Moreover, the limited energy supply of massive IoT is a chal-

lenging issue. Currently, most IoT devices are battery-powered

with small energy storage capacity. Thus, it is necessary to

replace the battery frequently to extend the lifetime of the

communication nodes. However, for massive IoT, frequent

battery replacement leads to a prohibitively high human cost

and environmental strain. In summary, massive access presents

many challenging unsolved issues, which cannot be addressed

with traditional approaches.

We note that the characteristics of massive access for

cellular IoT are very different from those of the other two

5G use cases, namely enhanced mobile broadband (eMBB)

and ultra-reliable low-latency communication (URLLC) [27].

In particular, eMBB aims to provide high data rates for broad-

band applications such as virtual reality (VR) or argument

reality (AR), while the objective of URLLC is to guarantee

ultra-reliable low-latency communications for critical missions

such as assisted/autonomous driving. Hence, for eMBB and

URLLC, OMA schemes are preferred to achieve high spectral

efficiency and link reliability. As pointed out above, 5G NB-

IoT also employs OMA, but as a result, it cannot fully realize

the goal of massive access [28]. For example, NB-IoT can only

accommodate fifty thousand devices per cell supporting a low

data rate [29]. For this reason, the ambitious goals of 5G NB-

IoT have to be realized by B5G cellular IoT. Without doubt,

the biggest challenge for B5G cellular IoT is the design of

effective multiple access schemes that meet the correspond-

ing performance requirements and services characteristics.

In Table II, we compare the performance requirements of

5G NB-IoT and B5G cellular IoT. Compared to 5G NB-

IoT, B5G cellular IoT imposes much more stringent require-

ments on power, connectivity, and coverage. Achieving these

performance requirements using traditional multiple access

techniques is very challenging. For example, it is challenging

to realize wide coverage with low transmit power. Hence,

new theoretical concepts, protocols, and techniques have to

be developed for B5G cellular IoT to realize massive access.

The research on B5G cellular IoT has already begun in

academia and industry. In [30], possible multiple access proto-

cols for B5G cellular IoT were surveyed, with a focus on grant-

free random access protocols based on approximate message

passing (AMP) algorithms. Massive multiple-input multiple-

output (MIMO) techniques for supporting cellular IoT were

reviewed in [31], and corresponding research opportunities

and challenges were identified. Moreover, as a promising

approach for B5G cellular IoT, non-orthogonal multiple access

(NOMA) was discussed in detail in [32]. A common viewpoint

of previous research is that B5G cellular IoT should further

exploit degrees of freedom in the spatial, frequency, and user

domains to facilitate significant performance improvements

[33]-[35]. Generally speaking, previous survey papers have

focused on one particular perspective of massive access, but

do not provide a comprehensive overview of massive access

in B5G cellular IoT which requires the consideration of many

different aspects. To accelerate the development of massive

access for the forthcoming B5G wireless networks, a com-

prehensive survey of the existing results, which can serve as

building blocks for new research on next-generation cellular

IoT, is necessary.

The objective of this paper is to provide such a compre-

hensive overview of the latest results and progress on massive

access in B5G wireless networks, c.f. Fig. 1. The remainder

of this paper is organized as follows. Section II introduces

3

Mas

sive

Acc

ess

Access Theory

Access Protocol

Access

Technique

Coverage

Enhancement

Grant-based

Random Access

Grant-Free

Random Access

CS-based

Detection

Covariance-

based Detection

Optimization

Algorithms

Bayesian

Approaches

Orthogonal

Access

Non-Orthogonal

Access

Outdoor

Indoor

Rural

Massive Random

Access

Massive Short-

Packet Access

Massive MIMO

mmW/THz

PD-NOMA

CD-NOMA

LDS

SCMA

Related Topics

Energy Supply

Access Security

Unsourced

Random Access

Sensing Matrix

Design

MUSA

Greedy

Algorithms

Fig. 1. Illustration of the aspects of massive access in B5G wireless networks considered in this survey paper. The labels of the rectangles correspond to the(sub)section titles.

TABLE IICOMPARISON OF 5G NB-IOT AND B5G CELLULAR IOT [36].

5G NB-IoT B5G Cellular IoT

Connectivity 50 thousand per cell 10 million per km2

Battery life 10 years 20 years

Coverage Ground Space-air-ground-sea

Latency 1 ms 0.3 ms

Reliability 10−4 10−6

Positioning 100 m 1 m for outdoor and 10 cm for indoor

massive access in cellular IoT. Then, Section III investigates

massive access from the perspective of information theory.

Massive access protocols, massive access techniques, and mas-

sive coverage enhancement are discussed in Sections IV, V and

VI, respectively. Moreover, energy supply for massive access

and massive access security are considered in Section VII.

Furthermore, future potential research directions for massive

access are described in Section VIII. Finally, Section IX

concludes the paper.

II. MASSIVE ACCESS IN CELLULAR IOT

Wireless access refers to the last-mile connection between

distributed end devices and a central station (e.g., a BTS).

Due to the limited radio spectrum, multiple wireless devices

have to share the same bandwidth employing multiple access

techniques. In general, the performance of multiple access is

determined by various factors such as channel conditions and

device requirements. First, the wireless channel may experi-

ence fading, interference, and noise, which can significantly

affect both access efficiency and reliability. Second, since mul-

tiple access schemes have to coordinate multiple devices, the

4

Fig. 2. Cellular IoT based on B5G wireless networks will be applied invarious fields, e.g., industry, agriculture, traffic, and medicine.

quality-of-service (QoS) requirements of the devices, e.g., rate

and latency, affect the selection of suitable access protocols

and techniques. Hence, the design of multiple access is always

a nontrivial issue. In B5G cellular IoT, c.f. Fig. 2, the evolution

from multiple access to massive access is driven by not only

the envisioned massive number of IoT devices, but also the

following critical service characteristics [37], [38]:

• Sporadic traffic: IoT devices do not always have data to

transmit creating bursty wireless traffic. In order to save

energy, idle devices do not access the network. In general,

a random number of devices access the network in each

time slot.

• Small payload: Most IoT applications infrequently gener-

ate small volumes of data having different sizes. In order

to improve the resource utilization efficiency, short-packet

transmission is preferable.

• Low power: Ideally, the batteries of IoT devices should

last for more than 20 years. Therefore, IoT devices have

to employ an intelligent transmit power strategy to reduce

the power consumption.

• Ubiquitous distribution: In order to support various ap-

plications, IoT devices are distributed over a wide range,

not only in urban areas, but also in rural areas. Hence,

wide wireless coverage is needed.

• Limited capability: Most IoT devices are wireless nodes

with simple architecture, and limited/no energy storage.

In other words, IoT devices cannot afford sophisticated

signal processing operations.

• Stringent latency constraint: Some IoT applications im-

pose stringent latency requirements. Low-latency access

schemes are needed to satisfy the latency requirements

of such IoT applications.

• Heterogenous QoS requirements: IoT devices across dif-

ferent application fields are very heterogeneous. For

example, a small sensor for temperature sensing and

a vehicle in a smart traffic system have very different

QoS requirements, leading to different requirements for

wireless access.

In general, massive access in B5G cellular IoT requires

low power, massive connectivity, and broad coverage. Yet,

the wireless channels for the last-mile connection between

distributed end devices and the central station constitute a

1R

2R

( )2C P

2

11

PCP

æ öç ÷+è ø

1

21

PCP

æ öç ÷+è ø

( )1C P

Fig. 3. The capacity region of the two-transmitter MAC.

major bottleneck in meeting these performance requirements.

First, the spectrum available in current wireless networks is

limited. Second, the coherence time of wireless channels is

also limited. For mobile applications such as smart traffic, the

coherence time becomes very short [39]. The length of the

coherence time constrains the length of a data frame, which

limits the performance of massive access. Moreover, for short

coherence times, it is difficult to obtain full channel state

information (CSI) for massive IoT and in some extreme sce-

narios with high mobility, CSI may not be available at all. In

such cases, non-coherent transmission which does not require

CSI may be adopted [40]. However, non-coherent transmission

suffers from a performance degradation compared to ideal

coherent transmission [41]. In summary, its unique character-

istics and the properties of the underlying wireless channels

lead to many challenging issues for realizing massive access.

In the following sections, we introduce potential solutions

from the perspectives of theories, techniques, and coverage

enhancement.

III. MASSIVE ACCESS THEORIES

Information theory is the foundation of modern commu-

nications and can provide useful guidelines for the design

of emerging wireless communication systems. To embrace

the challenges introduced by massive access, we first revisit

the capacity of the classical multiple access channel (MAC).

For the conventional MAC, the channel capacity has been

extensively studied [42]-[44]. It has been proven that the

capacity region of a K-transmitter MAC with unit channel

gain from each transmitter to the receiver can be characterized

by [45]

∑

k∈SRk < C

(

∑

k∈SPk

)

, ∀S ⊆ {1, · · · ,K}, (1)

where Rk and Pk are the kth transmitter’s data rate and

transmit power, respectively. The variance of the noise is

normalized to 1 and C(x) = log2(1 + x) is the Shannon

capacity formula. In Fig. 3, we plot the capacity region of

the two-transmitter MAC. The MAC capacity can be achieved

by performing successive interference cancelation (SIC) at

5

the receiver. For example, to achieve one corner point of

the capacity region, the receiver first decodes the signal of

one transmitter treating the signals of all other transmitters

as noise and removes the decoded signal from the received

signal. Then, the receiver decodes the next transmitter’s signal

treating the signals of the remaining transmitters as noise

and removes the decoded signal again, until all signals are

recovered. Note that for the MAC, the average per-transmitter

channel capacity, i.e., 1K

∑K

k=1 Rk, asymptotically approaches

zero as the number of transmitters K tends to infinity. This

is because co-channel interference becomes dominant when

there is a massive number of transmitters. The conventional

MAC capacity theory is only applicable for a fixed and finite

number of transmitters. In fact, the above MAC capacity region

is derived assuming infinitely long codes requiring a very large

number of channel uses, which is incompatible with the typical

IoT use cases [45]. Considering the characteristics of B5G

cellular IoT, an information theoretical study of massive access

has to take into account the following requirements:

• Massive connectivity: There is a massive number of IoT

devices with a density of more than 10 million devices

per km2.

• Random access: The sporadic traffic generated by typical

IoT applications leads to random activity of the devices.

Only active devices request access to the cellular network.

• Short-packet transmission: The small payload of IoT

data requires short-packet transmission to achieve high

resource efficiency and low access latency.

Hence, realizing massive access in practical systems re-

quires the development of new information theoretic design

guidelines, which are quite different from the results available

for the conventional MAC. Recently, the capacity of the

massive access channel has been derived. In the following,

we review these primary results.

A. Massive Random Access

The capacity of the massive access channel was first studied

in [46] based on the new notion of the many-access channel

(MnAC). The MnAC model studies the scenario in which the

number of transmitters increases unboundedly with the block-

length of the applied forward error correction (FEC) codes,

both tending to infinity. Specifically, by applying random

coding at the transmitters and Feinstein’s threshold decoding

at the receiver, as long as the number of transmitters K grows

sublinearly with the coding blocklength M , under a maximum

power constraint P , each transmitter can send to the receiver

a message of length

v(M) =M

KC(KP ) (2)

bits with an arbitrarily small error probability if M is suffi-

ciently large. Note that v(M) in (2) is a symmetric capacity

since all transmitters achieve the same capacity. This result

addresses the limitation of the conventional MAC capacity

theory in analyzing the capacity of massive access for the case

when the number of transmitters and the blocklength both go

to infinity.

In [46], the symmetric capacity of the MnAC for the case of

known active transmitter information is derived. Equivalently,

this corresponds to the scenario where the transmitters are

always active. However, as mentioned above, the transmitters

in massive access are expected to be randomly active due to

their sporadic traffic. For the case of random activity, practical

decoding schemes adopted at the receiver have to involve two

stages. The first stage identifies the set of active transmitters

based on the superposition of their unique signatures (this

corresponds to the active device detection problem in grant-

free random access as will be discussed in Section IV). The

second stage decodes the messages of the identified active

transmitters. Intuitively, activity identification may lead to a

loss in channel capacity. In [47], it is proven that for the

MnAC with random activity, by using random coding at the

transmitters and maximum-likelihood decoding at the receiver,

if the number of transmitters K grows as fast as linearly

with the coding blocklength M , the symmetric capacity of

a transmitter is given by

w(M) =

(

M

αKC(αKP )−

H2(α)

α

)+

, (3)

where (x)+ is defined as the maximum of x and 0, 0 ≤ α ≤ 1is the probability that a transmitter is active, and H2(α) =

−α ln(α) − (1 − α) ln(1 − α). Note that the termH2(α)

αis

the difference between the MnAC capacity with and without

activity information. Hence, the cost of activity identification

is equal to the entropy of the activity probability [48].

The above papers considered the case where all nodes are

single-antenna devices. For massive access in B5G wireless

networks, both the BTS and the IoT devices may employ

multiple antennas for performance enhancement [49], [50].

Specifically, deploying multiple antennas at the BTS is gen-

erally affordable and has become standard in modern com-

munication systems. Hence, it is necessary to characterize the

capacity of the MnAC for the multiple-input multiple-output

(MIMO) case. It was shown in [51] that when the number

of transmitters grows unbounded with the coding blocklength,

the asymmetric ergodic message-length capacity of transmitter

k is given by

uk(M) = ckEH

{

log2 det

(

INR+∑

t∈AHtQtH

†t

)}

−µkKH2(α) (4)

where H† is the conjugate transpose of H, EH{x} denotes

expectation with respect to random variable H, det(·) returns

the determinant of an input matrix, and A is the set of active

transmitters. Here, Ht ∈ CNR×NT and Qt ∈ CNT×NT are

the channel matrix from the tth transmitter to the receiver

and the covariance matrix of the codeword, respectively,

ck = limM→∞

Mµk, and µk = logLk∑

t∈A

logLt, where NR is the

number of antennas at the receiver, NT is the number of

antennas at each transmitter, and Lk is the number of messages

of the kth transmitter. The first term on the right hand side of

(4) is the individual capacity of the kth transmitter if activity

information is available at the receiver. Hence, the individual

6

capacity is proportional to the sum capacity with a scaling

factor ck, which depends on the number of messages. The

second term on the right hand side of (4) is the cost of activity

identification which is independent of the numbers of antennas

at the transmitters and the receiver.

B. Massive Short-Packet Access

The results in [46]-[48] and [51] on massive random ac-

cess are based on the common assumption that the coding

blocklength or the number of channel uses increases in the

same order as the number of transmitters. In this case, the

packet error rate (PER) can approach zero as the coding

blocklength tends to infinity. However, in practical systems,

the coding blocklength is finite. Especially, for cellular IoT,

short packets are preferred due to their lower latency for bursty

data communications [52], [53]. In the context of short-packet

transmission, it is difficult to guarantee error-free reception

for a short activity period. Thus, for a given packet length,

short-packet transmission should achieve a balance between

spectral efficiency and decoding error probability. Formally,

the capacity of short-packet transmission, R∗(M, ǫ, P ), can be

defined as the largest rate (log2 L)/M for which there exists

an (M, ǫ, P ) code, namely [54]

R∗(M, ǫ, P ) , sup

{

log2 L

M: ∃(M, ǫ, P )code

}

, (5)

where ǫ > 0 is the PER and L is the number of messages.

Via asymptotic analysis, it can be shown that (5) general-

izes the well-known existing capacity results. For instance,

as M → ∞, it is equivalent to Shannon’s capacity [55].

Moreover, when P tends to infinity, it is possible to obtain

the diversity-multiplexing tradeoff proposed by Zheng and Tse

[56]. However, it is challenging to derive the exact capacity

for massive short-packet access from (5) in closed form.

For the ease of analysis, a tight approximation for

R∗(M, ǫ, P ) was derived in [57] as follows

R∗(M, ǫ, P ) ≈ C(P )−

√

V

M

Q−1(ǫ)

ln 2, (6)

where Q−1(x) is the inverse Gaussian Q function, Q(x) =∫∞x

1√2π

exp(

− t2

2

)

dt, and V is the channel dispersion. In-

tuitively, as M tends to infinity, (6) reduces to the Shan-

non capacity formula. Based on (6), one can evaluate the

performance of massive short-packet access. Specifically, by

substituting the signal-to-interference-plus-noise ratio (SINR)

for massive short-packet access into (6), the achievable rate

for each transmitter can be evaluated.

Generally speaking, the results available for the capacity

of massive access in practical wireless networks are still

very limited. Most existing theoretical works only consider

Gaussian channels. If the channels suffer from fading and

need to be estimated, the capacities in (2) and (3) for mas-

sive random access will be quite different. Furthermore, for

massive short-packet access over fading channels, the capacity

of short-packet transmission essentially reduces to the outage

capacity [54]. A summary of existing results on massive access

information theory is given in Table III.

Active Device

Inactive Device

Fig. 4. Illustration of sporadic traffic of IoT applications. In general, duringan arbitrary time slot, only a fraction of the devices has data to transmit,namely the active devices.

IV. MASSIVE ACCESS PROTOCOLS

Due to the sporadic traffic of IoT applications, only a

fraction of the devices, namely the active devices, have data to

transmit at a given time, as shown in Fig. 4. Access protocols

are used to coordinate the access requests of the active IoT

devices [58]. Specifically, each active device contacts the BTS

to access the network. Then, the BTS identifies the active

devices by some means. Hence, an access protocol is needed

to coordinate the data exchange between the BTS and the IoT

devices. In general, the activity of the IoT devices is random.

Consequently, random access protocols are commonly used in

cellular IoT [59], [60]. Two types of random access protocols

are commonly distinguished, namely grant-based and grant-

free random access protocols [30]. Moreover, a new random

access protocol called unsourced massive random access has

been proposed recently [61]. We discuss these three massive

random access protocols in the following.

A. Grant-Based Random Access

Device BS

Fig. 5. Grant-based random access protocol.

Grant-based random access is adopted in the current 5G

NB-IoT [30]. As the name implies, for a grant-based random

7

TABLE IIISUMMARY OF INFORMATION THEORETICAL RESULTS FOR MASSIVE ACCESS SYSTEMS.

Reference System model Results

X. Chen et al. [46] Massive access with known activity, long packet, single antenna v(M) = MK

C(KP )

X. Chen et al. [47] Massive access with unknown activity, long packet, single antenna w(M) =(

MαK

C(αKP ) − H2(α)α

)+

W. Fan et al. [51] Massive access with unknown activity, long packet, multiple antennas uk(M) = ckEH

{

log2 det(

INR+

∑

t∈A HtQtH†t

)}

− µkKH2(α)

G. Durisi et al. [54] Multiple access with known activity, short packet, single antenna R∗(M, ǫ, P ) , sup{

log2 L

M: ∃(M, ǫ, P )code

}

G. Ozcan et al. [57] Multiple access with known activity, short packet, single antenna R∗(M, ǫ, P ) ≈ C(P ) −√

VM

Q−1(ǫ)ln 2

access protocol, an active device needs to obtain permission

from the BTS to access the network. As shown in Fig. 5, the

grant procedure of a typical grant-based random access pro-

tocol, such as ALOHA, includes four transmissions between

the IoT device and the BTS as follows [62]:

(1) Each active device randomly selects a preamble (also

referred to as a signature) from a pool of orthogonal

preamble sequences and uses the selected preamble to

inform the BTS that it has data to transmit.

(2) The BTS responds to each active device authorizing it to

send a connection request in the next stage.

(3) The active devices send connection requests for resource

allocation for data transmission.

(4) If a preamble is picked by only one active device, the

BTS will grant the corresponding request and send a

contention-resolution message to inform the active device

about the allocated resources. Otherwise, the access re-

quest is not granted.

The main advantage of the grant-based random access

protocol is the simple processing at the BTS. However, in

the context of massive access, the grant-based random access

protocol has the following shortcomings. First, the number

of orthogonal preamble sequences is finite due to the short

coherence time. If there exist a massive number of IoT

devices, the probability that a preamble is selected by more

than one device is high. In other words, the devices suffer

from a high probability of access failure due to collision.

As a consequence, the average access latency may become

too high to be tolerable. Second, the grant-based random

access protocol requires four transmissions, resulting in a high

signaling overhead. Since the channel capacity is limited, the

required signaling overhead might be prohibitively large for

massive access.

B. Grant-Free Random Access

To overcome the problems of grant-based random access,

various grant-free random access protocols have been pro-

posed which allow the active devices to access the wireless

network without a grant [63]. To be specific, active devices

first send their unique preambles to the BTS and then transmit

the data signals directly [64]. Therefore, both access latency

and signaling overhead are significantly reduced. The key idea

of grant-free random access is to detect the active devices

based on the received preambles at the BTS [65]. For massive

access, due to the massive number of devices and the use

of short packets, the preamble sequences are not orthogonal.

As a result, the received preamble signals suffer from severe

co-channel interference. Hence, the BTS has to adopt sophis-

ticated activity detection algorithms. In other words, grant-

free random access reduces the access delay and the signaling

overhead at the expense of a high computational complexity

at the BTS. In general, grant-free random access for massive

access requires massive device detection at the receiver. This

can be done using a compressed sensing (CS)-based sparse

signal recovery framework or a covariance-based approach for

massive device detection as discussed below.

1) CS Formulation: Due to the sporadic traffic generated

by IoT applications, the received preamble signal is typically

sparse when only the active devices send their preambles. It is

well known that the resulting sparse signal recovery problem

from noisy measurements can be tackled with CS methods

[66], [67]. The CS problem for massive device detection can

be formulated as

minimizeX

‖X‖0

s.t. ‖Y −AX‖F ≤ δ, (7)

where ‖·‖0 is the zero-norm defined as the number of nonzero

elements of the argument and ‖ · ‖F is the Frobenius norm. In

the above CS problem, Y is the space-time received preamble

signal and δ is a predetermined error tolerance constant which

depends on the noise power. Moreover, A = [a1, · · · , aK ] is

the sensing matrix with ak being the kth device’s preamble

sequence, and X = [α1h1, · · · , αKhK ]T is the device state

matrix with αi and hi being the activity indicator and the

channel response of the ith device, respectively. Here, αi = 1if the ith device is active, otherwise αi = 0. Hence, multiple

rows of X are zero and the aim of the CS problem (7) is to

determine the nonzero rows of X from the noisy measurements

Y, i.e., activity detection. The CS problem (7) is generally

nonconvex and thus it is difficult to obtain the globally optimal

solution directly and efficiently. Therefore, massive device

detection algorithms are usually designed based on a relaxed

CS problem. In what follows, we discuss different aspects of

the design of CS-based massive device detection algorithms.

[a] Sensing Matrix Design: The sensing matrix A has a

significant impact on the design of massive device detection

algorithms and hence determines the performance of grant-

free random access, namely the detection probability. Since

the preamble sequences are nonorthogonal, the design of the

sensing matrix is not a trivial task. In 4G LTE systems,

Zadoff-Chu (ZC) sequences are adopted as preamble signals

because of their good auto- and cross-correlation properties.

ZC sequences were used as preamble sequences for grant-

free random access systems in [68] and their performance

8

TABLE IVPREAMBLE SEQUENCES FOR MASSIVE DEVICE DETECTION.

Reference Preamble Sequence Advantage

J. Ding et al. [68] ZC sequences Good auto- and cross-correlation properties

L. Liu et al. [69] Gaussian sequences Easy to generate and convenient for performance analysis

J. Wang et al. [71] RM sequences Reduced storage space requirements

S. Li et al. [72] Deep auto-encoded sequences Adaptive to sparse patterns even without analytical models

was compared to that of orthogonal sequences. Recently, inde-

pendent and identically distributed (i.i.d.) Gaussian sequences

have been considered as preamble sequences to study grant-

free random access. This is because Gaussian sequences can be

easily generated and are convenient for performance analysis.

For instance, the detection probability for Gaussian distributed

preamble sequences was derived in [67] and [69]. It was shown

that the detection probability improved as the number of BTS

antennas was increased. Theoretically, the active devices can

be detected perfectly in the asymptotic limit as the number

of BTS antennas goes to infinity. Furthermore, the impact

of the length of Gaussian distributed preamble sequences on

the detection probability was analyzed in [70]. Then, the

durations of the preamble and data sequences in a frame were

optimized to maximize the system spectral efficiency. Since

each device is assigned a unique preamble sequence, the BTS

has to allocate a large amount of storage capacity to store

the preamble sequences in the massive access case. In order

to reduce the required storage space, Reed-Muller (RM) se-

quences can be applied as preamble sequences. The authors in

[71] exploited the nested structure of RM sequences and their

sub-sequences to design a low-complexity activity detection

algorithm. Moreover, a data-driven deep learning method was

applied to generate preamble sequences, which can adapt to

wireless channels with arbitrary distribution. In [72], a deep

auto-encoder was utilized to jointly design preamble sequences

and the corresponding sparse signal recovery algorithm, which

can effectively exploit sparsity patterns even without analytical

models. Several potential preamble sequences for massive

device detection are compared in Table IV.

[b] CS Algorithms: Since the zero-norm in the objective

function of the CS optimization problem (7) is nonconvex, it

is impossible to design massive device detection algorithms by

solving the original problem optimally with polynomial time

computational complexity [73]. In the literature, optimization

algorithms, greedy algorithms, and Bayesian approaches have

been utilized to obtain effective suboptimal solutions for the

above CS optimization problem in the context of massive

device detection. These algorithms are explained in the fol-

lowing.

[b.1] Optimization Algorithms: In order to obtain a feasible

solution of the CS problem (7), it is necessary to approximate

the objective function. In [74], the zero-norm ‖X‖0 was

replaced by the sum of all entries of X. Thus, the original

problem was transformed to a linear programming problem

which can be solved optimally with low complexity. The

solution of the linear programming problem can be proved

to be identical to that of the original problem only when

X is sufficiently sparse and there is no noise. On the other

hand, since the l1-norm is convex, many papers base the

algorithm design on l1-regularization problems. For example,

the authors in [75] proved that if X is sufficiently sparse,

l1-regularization problems can accurately recover X even

for noisy measurements. Furthermore, the authors in [76]

transformed the original CS problem to an l1-regularization

least-squares problem and proposed a customized interior-

point method for solving the problem.

For massive access in B5G wireless networks, the num-

ber of IoT devices and the number of BTS antennas are

expected to be very large, resulting in a high dimensional

device state matrix X. Hence, even with l1-regularization,

the computational complexity may still be prohibitive. In

fact, due to the spatial correlation of the BTS antennas, X

may not only be sparse, but also low-rank. The simultaneous

sparsity and low-rank property can be exploited to further

decrease the computational complexity of activity detection.

In [77], a rank or nuclear norm constraint was inserted into

the l1-regularization problem. Theoretical analysis revealed

that such a nuclear norm constrained problem can achieve

near-optimality based on a small number of measurements.

To further decrease the computational complexity and the

required length of the preamble sequences, a rank-aware l1-

regularization least-squares problem was formulated by esti-

mating the rank of X in advance [78]. For a given rank, the l1-

regularization least-squares problem can be transformed into

a low-dimensional problem. Yet, the rank-aware problem is

usually nonconvex. To tackle this challenge, a Riemannian

optimization-based algorithm was proposed in [78] to obtain

a suboptimal solution.

Moreover, lp-norm minimization with 0 < p < 1 can be

used to develop optimization algorithms for massive device

detection [73]. Since the lp-norm is a better approximation

for the l0-norm than the l1-norm, lp-norm minimization may

achieve more accurate activity detection. However, due to the

nonconvexity of the lp-norm, lp-norm minimization usually

leads to a high computational complexity.

[b.2] Greedy Algorithms: To avoid having to solve non-

convex CS problems via non-polynomial time algorithms,

greedy algorithms can be applied to massive device detection

in an effort to reduce the computational complexity at the

expense of a degradation in performance. Greedy algorithms

are iterative approaches that take local optimal decisions in

each step to eventually obtain an effective suboptimal solution.

One of the most widely used greedy algorithms in device

detection is the group orthogonal matching pursuit (GOMP)

algorithm [79]. Since the device state matrix X in general

contains only a few nonzero rows, in the absence of noise,

the measurements are in a space supported by a sub-matrix

of A determined by the positions of these nonzero rows. The

GOMP iteratively builds the support of X and enhances the

9

device detection performance by accumulating the correlation

between the residual and the sensing matrix. Although the

GOMP detection algorithm based on correlation is simple,

its performance is heavily affected by noise. Another type of

greedy algorithm, namely the Hierarchical Hard Thresholding

Pursuit (HiHTP) algorithm [80], [81], makes use of not only

the block structure of the device state matrix, but also the

intra-block sparse structure caused by the channel taps. The

sporadic device activity and the sparse channel profiles give

rise to a hierarchically sparse structured vector containing all

estimated channel coefficients. Motivated by this observation,

a prediction of the support of the device state matrix can

be inferred by applying a thresholding operation based on

the hierarchically sparse structure. Then, the best l2-norm

approximation to the received signal compatible with this

support is calculated. The HiHTP algorithm can efficiently

reconstruct hierarchically sparse signals from only a small

number of linear measurements.

One important property of greedy detection algorithms is

their simplicity of implementation. However, a drawback is

their inherent error propagation, since previous choices for

device activity are not re-evaluated. Moreover, the detection

performance of these algorithms is seriously affected by the

noise level.

[b.3] Bayesian Approaches: To improve the detection per-

formance of massive random access, Bayesian CS-based de-

tection algorithms have been developed, e.g. [82]-[85]. This

kind of detection algorithm first assigns a prior probability

distribution which promotes sparsity to the unknown device

state matrix, and then infers the posterior distribution of the

unknown signal from the received signal at the BTS. By

exploiting the prior channel information regarding the path loss

and the chunk sparsity structure, the authors in [82] proposed

a Bayesian CS-based algorithm to efficiently detect device

activity in an uplink cloud radio access network. However,

the Bayesian formulation in [82] was developed based on the

assumption of infinite-capacity fronthaul links, which is not

practical. Thus, taking into account the impact of fronthaul

capacity limitations, the authors in [83] employed a hybrid

generalized approximate message passing (GAMP) method,

which was based on a quadratic approximation of the sum-

product message passing scheme and accommodated both

nonlinear measurements and group sparsity to enhance the

device detection performance.

To further exploit statistical channel knowledge, the authors

in [67] adopted a Bayesian approach where the sparsity was

modeled via the prior distribution of the channel to facilitate

the development of an improved version of the approximate

message passing (AMP) algorithm. The authors in [69] further

demonstrated that in an asymptotic regime where the number

of users, the pilot length and the number of BTS antennas all

go to infinity in a particular manner, both the miss detection

and the false alarm probabilities of the AMP algorithm for

activity detection can asymptotically approach zero. The exact

knowledge of the prior distribution of the channels and the

noise variance may be difficult to obtain in practice due to

the sporadic traffic and the spatial correlation of the channels.

Furthermore, the above works considered massive device

detection in narrowband scenarios. In fact, B5G wireless

networks might employ broadband systems, e.g., millimeter

wave (mmW) or even terahertz (THz) systems [84]. Towards

this end, the authors in [85] proposed a generalized multiple

measurement vector approximate message passing (GMMV-

AMP) algorithm to adaptively detect the active devices by

exploiting the virtual angular domain sparsity of the channels

in an orthogonal frequency division multiplexing (OFDM)

broadband system. Furthermore, the expectation maximization

(EM) algorithm was utilized to learn the unknown hyper-

parameters of the channel and noise distributions. Several

massive device detection algorithms are compared in Table

V.

[c] Joint Device Detection and Channel Estimation: To

realize effective massive access, the BTS requires accurate

CSI for decoding the uplink signals and performing precoding

of the downlink signals after activity detection. In general,

CSI is acquired through channel estimation at the BTS based

on pilot sequences sent by the devices. Since the preamble

sequences for activity detection can be also exploited as

pilot sequences for channel estimation, activity detection and

channel estimation can be jointly performed based on the same

sequences.

Recently, several joint activity detection and channel estima-

tion (JADCE) algorithms for massive connectivity in cellular

IoT networks have been reported. Since JADCE is still a CS

problem, the authors in [86] proposed a Lasso-based l2,1-

regularization penalty function to exploit the inherent sparsity

existing in both the device activity and the remote radio

heads with which the active devices are associated. Then, an

alternating direction method of multipliers (ADMM) algorithm

was applied to handle the resulting large-scale convex JADCE

problem. In fact, Bayesian algorithms can be also adopted to

handle the JADCE problem. In [87], by exploiting statistical

information about the wireless channels, a JADCE algorithm

was designed for massive MIMO systems to jointly detect

device activity and estimate the CSI. Furthermore, an expec-

tation propagation (EP)-based JADCE algorithm was proposed

in [88] for massive access. This algorithm approximated

the computationally intractable probability distribution of the

sparse channel vector by an easily tractable distribution, which

can substantially enhance JADCE performance.

A major problem of the above JADCE algorithms is that

in practical scenarios with short pilot sequences, their per-

formance is severely degraded. To tackle this problem, the

authors in [89] proposed a transmission control scheme for

grant-free random access protocols. Specifically, based on

a predetermined transmission control function, each active

device decides to transmit a packet in the current time slot

or to postpone the transmission. At the BTS, a modified

AMP algorithm was adopted to improve JADCE performance.

Transmission control was motivated by the fact that decreasing

the number of active devices can significantly improve JADCE

performance for a given pilot length.

[d] Joint Device and Data Detection: To reduce access

latency and signaling overhead, blind detection has become

a promising approach to jointly detecting devices and data for

massive access scenarios without prior knowledge of the CSI,

10

TABLE VMASSIVE DEVICE DETECTION ALGORITHMS.

Reference Algorithm Type Description

M. Golbabaee et al. [77] Optimization algorithm Rank-constrained l1-regularization optimization algorithm

X. Shao et al. [78] Optimization algorithm Rank-aware Riemannian optimization algorithm

C. Bockelmann et al. [79] Greedy algorithm Group orthogonal matching pursuit algorithm

I. Roth et al. [80] Greedy algorithm Hierarchical hard thresholding pursuit algorithm

Z. Chen et al. [67] Bayesian algorithm Approximate message passing algorithm

M. Ke et al. [85] Bayesian algorithm Generalized multiple measurement vector approximate message passing algorithm

especially for low-latency communications. For instance, the

authors in [66] proposed a non-coherent transmission scheme

that does not need CSI at the BTS and developed a modified

AMP algorithm to exploit the structured sparsity caused by

the scheme. For this algorithm, explicit channel estimation is

not required because of the non-coherent transmission, and the

data signal is embedded into the pilot sequences. Motivated

by the observation that if the active devices transmit symbols

that are either −1 or 1 and the inactive devices are modelled

as transmitting all-zero symbols, the transmit symbol alpha-

bet is ternary, the authors in [90] proposed an information-

enhanced adaptive matching pursuit algorithm for joint device

and data detection. Moreover, the authors in [91] proposed

a maximum a posteriori probability (MAP)-based device and

data detection algorithm, which comprises a MAP-based active

user detector (MAP-AUD) and a MAP-based data detector

(MAP-DD). Extrinsic information is exchanged between the

MAP-AUD and the MAP-DD. In particular, joint detection

of the active devices and the data symbols is performed first,

then the estimated data symbol is refined and used as a priori

information for the detection of the active devices.

The above algorithms carry out joint device and data

detection within one time slot. In other words, they do not

exploit the temporal correlation across time slots. To this

end, a dynamic CS-based device and data detector using

orthogonal matching pursuit (OMP) across time slots was

proposed in [92]. Moreover, an a priori information aided

adaptive subspace pursuit (PIA-ASP) algorithm was proposed

in [93] to detect active devices and data symbols. In the PIA-

ASP algorithm, a parameter evaluating the quality of the prior-

information support set was introduced, so as to exploit the

intrinsic temporal correlation of the active device support sets

across several continuous time slots.

2) Covariance Formulation: If we are only interested in

detecting the device activities and not interested in estimating

the channel and if the BTS is equipped with a large number

of antennas, it is possible to formulate the massive device

detection as a maximum likelihood estimation problem based

on the covariance matrix of the received signal at the BTS,

and then employ a coordinate descent method to obtain a

suboptimal solution [94], [95]. The key advantage of this

covariance-based approach is that it is able to detect a much

larger number of active devices. In fact, the number of active

devices can scale quadratically with the length of the pilot

sequences, thereby alleviating a key bottleneck in massive

access. This scaling law was established under a so-called non-

negative least square (NNLS) formulation in [95], and can also

be analyzed via the Fisher information matrix of the maximum

likelihood problem [96]. We note that the covariance-based

approach can also be used for joint activity and data detection.

Specifically, each device is assigned not only one sequence but

a unique sequence set. The transmitted sequence corresponds

to the transmitted data. Hence, by detecting the received

sequence, the activity information and the data information

can be obtained simultaneously.

Generally speaking, massive device detection for grant-

free random access is still an open problem, which involves

two challenging unsolved issues. First, the existing algorithms

entail a high computational complexity for recovering the

device state matrix due to its large-dimensional structure.

Second, the required length of the preamble sequences may

be too long for short-packet transmission in B5G wireless

networks.

C. Unsourced Random Access

Recently, a new massive random access paradigm, referred

to as unsourced massive random access, was proposed in

[61]. Unlike grant-based and grant-free random access pro-

tocols that assign each device a unique preamble sequence,

unsourced massive random access utilizes one codebook (a

set of sequences) for all devices. The devices include their

identity (ID) in the information message itself, and the BTS

decodes the list of active device messages up to permutations.

It has been shown that unsourced massive random access can

significantly decrease the minimum energy per bit required

for reliable communication. The authors in [97] extended

unsourced massive random access to the case where the BTS

had a very large number of antennas and no CSI. Specifically,

the minimum energy required for reliable communication can

be made arbitrarily small as the number of BTS antennas

grows sufficiently large. However, there are many challenging

unsolved problems in unsourced massive random access, such

as efficient codebook design and activity detection algorithms.

Some recent progress on codebook design for massive access

has been reported in [98].

V. MASSIVE ACCESS TECHNIQUES

Access techniques organize the data exchange between the

active devices and the BTS. In previous generations of cellular

networks, OMA techniques, such as TDMA, FDMA, and

OFDMA, have been adopted. For 5G NB-IoT, SC-FDMA is

employed for the uplink and OFDMA for the downlink. In par-

ticular, OMA techniques allocate each time-frequency resource

block to a unique device which leads to a simple transceiver

structure. However, due to the limited radio spectrum available

for cellular communications, it is difficult to support massive

11

access with the conventional OMA techniques. To tackle this

challenge, there are two possible directions for massive access

in B5G wireless networks. On the one hand, new wireless

resources, e.g., new spectrum, can be utilized to admit more

devices. On the other hand, resource utilization efficiency

can be further improved to support massive access. In the

following, we describe two different massive access techniques

which are along the above-mentioned directions.

A. Massive Orthogonal Access

Conventional OMA techniques over limited radio spectrum

cannot satisfy the stringent QoS requirements of massive

access, and hence B5G wireless networks have to adopt

new massive orthogonal access techniques by exploiting extra

degrees of freedom. Due to the strict latency constraints, time-

domain resources are scarce and cannot be used for massive

access. Instead, exploiting additional space- and frequency-

domain resources is more attractive.

1) Massive MIMO: Multiple-antenna techniques have been

adopted in 4G long-term evolution (LTE) networks to increase

transmission rate and to enhance link reliability by exploiting

extra spatial degrees of freedom. However, the BTSs of 4G

LTE can be equipped only with up to eight antennas. Thus, the

spatial degrees of freedom offered by LTE multiple-antenna

BTSs are limited and far from enough to facilitate massive

access. To significantly increase the available spatial degrees

of freedom, the BTSs of B5G wireless networks will deploy a

large-scale antenna array with 64 or more antennas, realizing

massive MIMO. Therefore, a large number of devices can

access the network simultaneously and the BTS can separate

them in the spatial domain, e.g. space division multiple access

(SDMA) [99], [100].

The pioneering work on massive MIMO in [101] showed

that co-channel interference vanishes asymptotically even with

simple linear precoders and combiners as the number of BTS

antennas tends to infinity due to channel hardening. Moreover,

it was shown that both the spectral and energy efficiencies

can be improved significantly by using massive MIMO [102],

[103]. In the case of massive access, massive MIMO does

not only increase the accuracy of active device detection but

also improves the transmission performance [87]. However,

there are two critical issues for implementing massive access

in massive MIMO systems.

The first issue concerns the CSI acquisition at the BTS. The

accuracy of the CSI at the BTS determines the performance

of massive access based on massive MIMO [104]. Since the

BTS is at the transmitter side for the downlink, it is impossible

to obtain CSI directly. In traditional multiple-antenna systems,

there are two CSI acquisition methods. In frequency division

duplex (FDD) systems, the devices first obtain the CSI by

channel estimation and then feed back the quantized CSI to

the BTS [105]. For massive access based on massive MIMO,

the required number of feedback bits at each device is large

due to the high dimensional channel vector. Thus, the total

amount of feedback required for a large number of devices

can be prohibitive. In other words, conventional quantized

feedback methods are not applicable for massive access based

on massive MIMO. However, if the channel is sparse, several

effective methods, e.g. CS [106], [107], can reduce the amount

of feedback. In [108], the received pilot signal of each device

was conveyed to the BTS and the sparse CSI of all devices

was jointly recovered by using a l1-regularization-based CS

method. Moreover, deep learning was employed to compress

the sparse CSI in [109], such that the amount of feedback

became affordable. The requirement of sparse CSI limits the

applicability of the above methods to FDD systems. Hence,

massive MIMO is usually envisioned to operate in the time

division duplex (TDD) mode. In TDD systems, the devices

send pilot sequences to the BTS in the uplink and the BTS

obtains the downlink CSI by estimating the uplink channels

exploiting channel reciprocity [110]. An obstacle to realiz-

ing CSI acquisition in TDD systems is the so-called pilot

contamination problem [111]. Specifically, due to the limited

pilot sequence length available for serving a massive number

of devices, the same pilot sequences have to be reused in

different devices. Consequently, the CSI estimation accuracy

is reduced due to co-channel interference. Because the pilot

sequences in massive access cannot be completely orthogonal,

it is necessary to improve the CSI accuracy. To this end, in

[112], a pilot transmit power control scheme was proposed, so

as to improve the overall performance. Since the CSI accuracy

depends on the pilot transmit energy, for a given transmit

power, the pilot sequence length should be optimized as in

[70].

The second issue concerns the energy consumption and

the associated cost. The use of massive MIMO for massive

access requires a large number of radio-frequency (RF) chains

and the associated analog-to-digital converter (ADC) modules.

If each antenna is equipped with a dedicated RF chain, the

number of RF chains can be very large resulting in high energy

consumption [113]. To decrease the number of RF chains

but retain the benefits of massive MIMO, hybrid precoding

techniques are needed to allow multiple antennas to share

the same RF chains [114]–[116]. Compared to conventional

digital precoding schemes, hybrid precoding incurs high de-

sign complexity but low implementation cost. In [117], a

penalty dual decomposition-based hybrid precoding design

method was proposed, which is guaranteed to converge to a

Karush-Kuhn-Tucker (KKT) solution of the hybrid precoding

problem under some mild assumptions. To decrease the design

complexity, a two-timescale hybrid precoding method was

presented in [118], which constructed the analog beamforming

based on slowly time-varying statistical CSI. Moreover, the

high cost of ADC is a vital issue for massive MIMO. Since

the ADC cost is mainly determined by the resolution of the

quantization, a low-resolution ADC is preferred for massive

access based on massive MIMO at the cost of a performance

loss [119], [120]. In [121], the impact of low-resolution ADC

on the performance of massive access was analyzed, and a time

allocation scheme for channel estimation and data transmission

was proposed to alleviate the impact of low-resolution ADC.

2) Millimeter-Wave/Terahertz: According to the Shannon

capacity theorem, increasing the bandwidth is a simple but

effective way for improving the capacity of wireless com-

munications. Current cellular networks operate in sub-6 GHz

12

bands which provide limited bandwidth and are overcrowded.

On the contrary, high frequency bands have large vacant

spectra. Hence, for B5G wireless networks, the use of mmW

and even THz bands is attractive in order to realize massive

access in the frequency domain [122]-[124]. A critical issue

for mmW/THz communications is the severe propagation

loss, resulting in a short transmission distance. To address

this problem, mmW/THz is usually combined with massive

MIMO or even ultra-massive MIMO employing more than one

thousand antennas [125]. Thus, CSI acquisition and precoding

become more complicated for mmW/THz communications.

Fortunately, mmW/THz channels have two important charac-

teristics. Firstly, the mmW/THz channel is very sparse, such

that CS and Bayesian methods can be used to acquire the CSI

required for the design of the precoding matrix. In [126], to

obtain CSI, the l1,2-regularization-based CS method was ap-

plied, which can avoid quantization errors and provide super-

resolution performance. By modeling the channel coefficients

as Laplacian distributed random variables, a GAMP algorithm

was used to find the entries of the unknown mmWave MIMO

channel matrix in [127]. Secondly, mmW/THz channels usu-

ally exhibit high-resolution angular-domain characteristics.

Accordingly, beam tracking methods can be used to extract

the CSI. In [128], a beam selection scheme was presented to

decrease the complexity of beam tracking. Moreover, beam

alignment was applied to improve the performance of beam

tracking in [129].

B. Massive Non-orthogonal Access

A promising approach for increasing the number of sup-

ported access devices over a limited radio spectrum is the

use of non-orthogonal multiple access (NOMA) techniques,

which allow multiple devices to share the same time-frequency

resource block. Hence, NOMA is a candidate technique

for B5G wireless networks [130]-[133]. Compared to OMA

techniques, NOMA techniques have the potential to improve

spectral efficiency. In other words, for a given required spectral

efficiency per device and a given bandwidth, NOMA can

admit significantly more devices than OMA. Thereby, NOMA

is able to support massive access in a limited radio spec-

trum. However, NOMA techniques lead to severe co-channel

interference, especially in the massive access scenario. The

key to realizing massive NOMA is interference management

[134]. So far, academia and industry have proposed several

NOMA schemes, which can be classified into two categories,

namely power-domain non-NOMA (PD-NOMA) and code-

domain NOMA (CD-NOMA).

1) Power-Domain Non-Orthogonal Multiple Access: PD-

NOMA shares the radio spectrum through superposition cod-

ing with the transmit powers as weight factors [135]. In this

case, the access devices can be separated in the power domain.

In order to decrease the co-channel interference caused by non-

orthogonal transmission, successive interference cancelation

(SIC) is usually carried out at the receiver. Specifically, the

receiver first decodes the interfering signal with the highest

transmit power and removes it from the received signal. Then,

it decodes the interfering signal with the next highest transmit

power until the desired signal is recovered. Hence, power

allocation has a great impact on the performance of PD-

NOMA. Intuitively, a device with a small channel gain is

allocated a high transmit power, so as to guarantee fairness.

Yet, it is not a trivial task to perform optimal power allocation

in PD-NOMA, since there is residual inter-user interference

from the devices with lower transmit powers. In [136], a

cognitive power allocation scheme was proposed for two-user

PD-NOMA. Since SIC only cancels the partial co-channel

interference caused by the devices with higher transmit pow-

ers, devices with small channel gains may still suffer from

strong co-channel interference after SIC, resulting in poor

performance. In order to guarantee fairness, the authors in

[137] proposed a power allocation scheme that maximizes the

rate of the device with the smallest channel gain.

PD-NOMA improves spectral efficiency at the cost of high

computational complexity due to the use of a SIC receiver

for interference mitigation. The computational complexity

increases as the number of devices increases. In the scenario of

massive access, the computational complexity and the signal

processing delay might be prohibitive if SIC is performed for

all devices. A possible solution to overcome these challenges

is to perform device clustering [138], [139]. Thereby, the

devices are grouped into several clusters, where each cluster

contains a small number of devices. SIC is performed within

each clusters, which reduces the computational complexity

effectively. In [140], a frequency-domain clustering scheme

was proposed, where two devices assigned to the same sub-

carrier of an OFDMA system form a cluster. Frequency-

domain clustering guarantees orthogonality across clusters,

but decreases the spectral efficiency. Considering that the

BTSs of B5G wireless networks will be equipped with a

large-scale antenna array, it may be preferable to perform

device clustering in the spatial domain, where each cluster

can occupy the entire spectrum [141]. Since spatial device

clustering incurs extra inter-cluster interference, spatial beam-

forming combined with power allocation has to be employed to

combat the interference [142]. To further unleash the potential

of non-orthogonal signaling, a fully non-orthogonal access

scheme was designed for massive access in [143], where non-

orthogonal pilot sequences were used to estimate the CSI with

small overhead, and the estimated CSI was applied for the

design of spatial beamforming for interference cancellation.

As mentioned earlier, most IoT devices are simple nodes

with limited computational capability. As a result, IoT devices

may perform SIC imperfectly, resulting in severe residual

intra-cluster interference. In this context, the authors in [144]

proposed a spatial beamforming scheme for massive access

taking imperfect SIC into consideration. In fact, the design

of a large number of spatial beams canceling inter-cluster

interference also entails high complexity in the massive access

scenario. In order to reduce the computational complexity

of beam design, a beamspace non-orthogonal multiple access

scheme was proposed in [145], which constructed the transmit

beams based on statistical CSI. Compared to the beamforming

schemes based on instantaneous CSI, the one based on statis-

tical CSI leads to a complexity reduction at the cost of a loss

in performance.

13

TABLE VIMASSIVE NON-ORTHOGONAL ACCESS SCHEMES.

Reference Type Characteristics

X. Chen et al. [143] PD-NOMA Superposition coding at the transmitter and SIC at the receiver

Y. Du et al. [147] LDS-CDMA Spreading of the transmitted symbols in the time domain by a low-density code and MAP at the receiver

R. Razavi et al. [148] LDS-OFDM Spreading of the transmitted symbols in the frequency domain by a low-density code and MAP at the receiver

Z. Yuan et al. [149] MUSA Spreading of the transmitted symbols by a code selected from a set of multiple sparse codes and MAP at the receiver

F. Wei et al. [152] SCMA Mapping of the transmitted symbols into a codeword of a codebook consisting of multiple sparse codes and AMP at the receiver

2) Code-Domain Non-Orthogonal Multiple Access: CD-

NOMA assigns different codes to devices for multiplexing

[146]. Different from conventional CDMA, the assigned codes

are sparse. However, the sparse codes can still offer spread-

ing gains for suppressing undesired co-channel interference.

Hence, only a simple message passing algorithm (MPA) at the

receivers is needed to detect the sparse CD-NOMA sequences.

Low-density signature CDMA (LDS-CDMA) [147] and low-

density spreading OFDM (LDS-OFDM) [148] are two direct

extensions of CD-NOMA. In particular, in LDS-CDMA, the

symbol to be transmitted is spread in the time domain, while

in LDS-OFDM, the chips are transmitted in the frequency

domain. LDS-CDMA and LDS-OFDM can be selected based

on the massive access system requirements.

In practice, LDS-CDMA and LDS-OFDM spread the signal

using a predetermined sparse code for each device. In fact,

if the device has multiple spreading codes, it is possible to

further improve the performance of massive access. Inspired

by this idea, multi-user shared access (MUSA) was proposed.

For MUSA, there is a set of spreading codes [149]. Each

device randomly selects a spreading code for each symbol, and

thus in fading channels, the average interference is suppressed

due to the use of different spreading codes at the cost of a

high-complexity receiver.

Unlike the above CD-NOMA schemes, sparse code multiple

access (SCMA) maps the symbol to a sparse code [150].

Each device has a predetermined codebook containing multiple

sparse codes, where the nonzero elements are in the same

positions. The symbol to be transmitted is mapped to an

index, and the corresponding sparse code in the codebook is

selected for transmission. The codebook design for SCMA

was discussed in [151] to further improve the performance

of SCMA. Considering the requirements of massive access,

an SCMA scheme with joint channel estimation and data

decoding was proposed in [152].

A comparison of different massive non-orthogonal access

schemes is provided in Table VI. In summary, both PD-NOMA

and CD-NOMA exploit new degrees of freedom for channel

sharing so as to support massive access over limited wireless

resources. However, both massive access techniques require

sophisticated transceivers to combat co-channel interference.

Considering that the channel matrices in massive access are

high dimensional due to the deployment of the large-scale

antenna arrays at the BTSs, the computational complexity

at the transceivers may be prohibitive. Therefore, the design

of simple but effective transceivers is an important topic for

future research.

CPU

AP

Fig. 6. A cell-free massive MIMO system, where multiple access pointsdistributed over the whole area connect to a central processing unit throughhigh-capacity optical fibre links. Thus, the access distances are shortenedsignificantly.

VI. MASSIVE COVERAGE ENHANCEMENT

IoT has found various applications in industry, agriculture,

traffic, medicine, etc. Hence, IoT devices are distributed over

a very large range, not only in urban areas, but also in rural

areas. To decrease the power consumption and achieve long

battery usage periods, e.g. 20 years, the transmit power of

IoT devices is typically smaller than 23 dBm. Therefore,

the signal received from cell-edge devices is usually very

weak, such that it is difficult to satisfy the QoS requirements.

Consequently, the coverage of current cellular IoT is limited.

Especially, signals received from indoor wireless devices are

usually weak, but there is a large number of such devices.

As a result, indoor massive access is a critical issue. 5G NB-

IoT adopts several coverage enhancement schemes, e.g., low-

order modulation and retransmission, to improve the quality

of signals originating from the cell-edge and indoors. These

schemes enhance the coverage at the cost of a low resource

utilization efficiency. Yet, for massive access, there is no extra

resource that can be used for coverage enhancement. More-

over, current cellular networks only cover densely populated

areas, but IoT has been applied also in rural areas. Deploying

new cellular networks in rural areas is inefficient in terms of

capital cost. Therefore, it is necessary to develop new coverage

enhancement strategies for massive access in B5G wireless

networks. In the following, we discuss three possible coverage

enhancement strategies.

A. Cell-Free Massive MIMO

Massive MIMO has been proved to be an effective approach

to enhance the coverage by exploiting its large array gain

14

Fig. 7. Indoor coverage enhancement with an intelligent reflecting surface,where the intelligent reflecting surface comprising a large number of reflectingunits generates a favorable propagation environment via beamforming and iscontrolled by a microcontroller.

[153]. Recently, it has been shown that cell-free massive

MIMO can further improve coverage performance. Cell-free

massive MIMO is indeed a distributed antenna system, which

comprises a large number of access points (AP) connected to

a central processing unit (CPU) [154], [155], as shown in Fig.

6. Each AP can deploy one or multiple antennas. The system

is not partitioned into cells and each user is served by one or

multiple APs. Compared to co-located massive MIMO, cell-

free massive MIMO significantly shortens the access distance,

and thus broadens the coverage area. Since the APs form a

large-scale antenna array, the same high spectral efficiency

as with conventional massive MIMO can be achieved. In

[156], it was proved that under uncorrelated shadow fading

conditions, cell-free massive MIMO provides a nearly five-

fold improvement in the 95%-likely per-user throughput over

a small-cell architecture, which is an enhancement strategy for

4G LTE [157], and a ten-fold improvement under correlated

shadow fading conditions. Thus, cell-free massive MIMO is

a promising choice for outdoor coverage enhancement at low

power.

In practice, the APs equipped with independent RF chains

are connected to the CPU by high-capacity optical fibres. Since

the devices are randomly distributed over the service area,

each AP has a different impact on the overall performance.

Hence, it is necessary to wisely allocate the wireless resources

to the APs to achieve the optimal system performance. A

max-min power control scheme was proposed in [158] to

provide equal throughput for all users. It was found that most

APs transmitted at less than the maximum possible power.

Moreover, to improve the utilization efficiency of the low-

resolution ADCs in cell-free massive MIMO, a quantization

bit allocation scheme was proposed in [159].

B. Intelligent Reflecting Surface

The deployment of IoT devices located indoors is expected

to increase significantly in the coming decade [160]. For

indoor applications, in general, the received signal is weak

due to the attenuation caused by walls [161]. Hence, indoor

coverage enhancement is crucial for realizing massive access

in indoor scenarios. Considering the difficult propagation

environment, indoor coverage enhancement is not a trivial

task. A possible solution to this problem is to control the

reflection characteristics of walls to establish favorable signal

propagation environments. The concept of an intelligent wall

as an autonomous part of a smart indoor environment was

proposed in [161]. In particular, an intelligent wall is a wall

equipped with an active frequency-selective surface, simple

low-cost sensors, and a cognitive engine, which can control

the radio coverage to improve the overall system performance.

A simple but effective way to realize such a wall is the

deployment of a reconfigurable reflect-array, also referred to

as an intelligent reflecting surface (IRS), with a large number

of reflecting units that reflect the transmitted signals [162], as

shown in Fig. 7.

By optimally controlling the phase shift of each unit of

the IRS, the desired signal can be enhanced while undesired

interference can be canceled [163], [164]. Since the IRS

comprises a large number of reflecting units, it can play

the same role as a large-scale antenna array through spatial

beamforming. As a result, even for a large number of indoor

wireless devices, the quality of the received signals can be

significantly improved. However, compared to the large-scale

antenna array, the IRS entails lower cost and complexity.

This is because the large IRS does not require power-hungry

RF chains. In other words, IRS is a green massive coverage

enhancement strategy. In [165], the IRS phase shifters were

optimized to maximize the sum rate in a multiuser scenario.

Taking into account that the number of phase shifts for each

reflecting unit is finite in practice, IRS beamforming was

optimized to minimize the total power consumption in [166].

It is worth pointing out that IRS can be utilized to enhance

not only indoor coverage, but also outdoor coverage. Hence,

IRS is regarded as a promising technique for B5G wireless

networks [167]-[169]. Exploiting IRS specifically for massive

access is an interesting topic for future work.

C. Satellite Communications

The rural deployment of IoT is important for monitoring and

management applications [170]. For instance, security cameras

have been installed in forests to predict wildfires [171], and

a large number of sensors have been deployed in the sea to

monitor the ocean resources [172]. Currently, these areas are

not covered by cellular networks. From a cost perspective, it

is prohibitively expensive to deploy new cellular networks in

rural areas. Thereby, satellite communications are expected to

provide wireless access in rural areas [173]. In particular, a

satellite can cover a large area, and thus the cost of wireless

access is substantially decreased. Hence, satellite communi-

cation is expected to become an important component of

B5G wireless networks [174], [175]. By integrating space and

ground networks, seamless coverage can be provided all over

the world.

To shorten the access latency, low earth orbit (LEO) satel-

lites are usually used as access points for space networks

[176]. By applying multiple-beam techniques, a LEO satellite

15

Fig. 8. Coverage enhancement by a multi-beam LEO satellite, which canprovide low-latency and reliable wireless access to a large number of devicesdistributed in large rural areas by using multiple spatial spot beams.

can serve a large number of devices simultaneously, as shown

in Fig. 8. In [177], the transmit beamforming was designed

for multicast in multiple-beam satellite communications. Con-

sidering imperfect CSI at the satellite, a robust beamfoming

scheme was developed in [178] with the objective of minimiz-

ing the total power consumption. Moreover, cooperative mul-

ticast transmission in integrated space-ground networks was

investigated in [179]. Furthermore, a max-min beamforming

scheme was designed to jointly optimize the beamforming

vectors of the BTS and the satellite. With the fast evolution

of high-throughput multiple-beam satellite communications,

satellite IoT has been proposed in [180], and is expected to

accelerate the creation of an IoE.

Massive access is subject to complex and time-varying

propagation environments. Hence, it is natural to enhance

wireless coverage by combining multiple strategies. For ex-

ample, in hotspot areas, both cell-free massive MIMO and

IRS can be employed to enhance the signal quality. However,

the combination of multiple enhancement strategies entails

high implementation complexity and cost, which have to be

carefully considered for practical deployment.

VII. OTHER MASSIVE ACCESS TOPICS

As outlined in Sections III-VI, massive access in B5G

wireless networks involves many aspects, including theoretical

concepts, protocol design, algorithm development, and cover-

age extension, which have to be jointly considered to improve

the efficiency and reliability of massive access. However, in

order to realize massive access in B5G wireless networks,

there are additional critical issues which have to be considered

such as energy supply and access security. Herein, we provide

a brief discussion of these two topics from the perspective of

massive access.

A. Wireless Energy Transfer

Currently, most IoT devices are battery powered. Since

the battery capacity of IoT devices is quite limited, the

transmit power has to be very low, e.g., 23 dBm. The low

transmit power limits the capabilities of IoT applications.

On the other hand, for higher transmit power, the battery

has to be replaced frequently. The battery replacement for

a massive number of IoT devices entails a high human cost

and a large environmental strain. Moreover, it is difficult to

replace the batteries of devices in extreme environments, e.g.,

in walls and under water. Recently, wireless energy transfer,

namely wireless charging, has received considerable attention

from both academia and industry [184]-[186]. In particular,

wireless energy transfer based on RF signals can provide stable

and reliable energy supply. Hence, IoT devices can realize

sustainable communications even under adverse conditions,

as long as there is wireless coverage. More importantly, due

to the broadcast nature of wireless channels, many devices

can be charged in parallel. Hence, wireless energy transfer

is particularly appealing for cellular IoT with massive access

[187], [188].

A challenging issue in wireless energy transfer is the low

energy transfer efficiency due to path loss and channel fading

during the transmission of the wireless energy signal. As a

result, the effective distance of wireless energy transfer is

too short to achieve the broad coverage desirable for massive

access. To overcome this problem, the concept of energy

beamforming was introduced for wireless energy transfer

[189], [190]. Specifically, by using spatial beamforming, the

energy signal is focused on the receiver, and thus the transfer

efficiency can be improved effectively. It was shown that even

with partial CSI at the transmitter, energy beamforming can

enhance the energy transfer efficiency. Especially in multiuser

scenarios, energy beamforming can facilitate simultaneous en-

ergy harvesting for a massive number of devices. Furthermore,

when the transmitter is equipped with a large-scale antenna

array, the effective transmission distance can be increased sig-

nificantly [191]. By exploiting the very high spatial resolution

of large-scale antenna arrays, it is possible to charge a massive

number of devices with high efficiency. Moreover, multiple-

point cooperation and relaying can be employed to further

increase the transfer distance [192]-[194].

Wireless energy transfer is already being applied for short-

distance charging scenarios. For instance, mobile phones can

be charged without wireline connection. The provision of long-

distance wireless energy transfer for practical massive access

is still an open research problem due to the low transfer

efficiency and requires further research.

B. Physical-Layer Security

In massive access, a massive number of devices share the

radio spectrum. Any device can receive the other devices’

signal, resulting in the risk of information leakage. Tradi-

tionally, access security has been realized by using upper-

layer encryption techniques [195], [196]. Due to the fast

evolution of communication technology, the computational ca-

pabilities of eavesdropping nodes have significantly increased.

Consequently, encryption techniques have to become more

sophisticated to guarantee information security. Yet, most

IoT devices are low-cost nodes with limited computational

capability, and thus they cannot afford the high complexity

required for advanced encryption techniques. Moreover, for

16

some versions of massive access, e.g., grant-free random

access, conventional encryption techniques relying on secure

key distribution are not applicable. In this context, physical-

layer security techniques, as a complement to conventional

encryption techniques, can be adopted to facilitate secure

massive access [197]. The essence of physical-layer security

is to exploit the inherent random characteristics of wireless

channels, e.g., fading, interference, and noise, to ensure that

the information transmission rate of the desired link is higher

than the eavesdropping channel capacity, and hence, the eaves-

dropper is not able to decode the intercepted signal correctly

[198], [199]. As mentioned earlier, massive access causes

severe co-channel interference, which can be exploited to

improve the security of B5G wireless networks by applying

physical-layer security techniques.

According to the basic principles of physical-layer security,

in order to enhance the secrecy performance, it is necessary

to improve the quality of the legitimate signal and decrease

the quality of the eavesdropping signal simultaneously. Hence,

multiple-antenna techniques are commonly employed to pro-

vide physical-layer security [200]. For instance, if the legiti-

mate signal is transmitted in the null space of the eavesdrop-

ping channel matrix, the eavesdropper cannot receive the le-

gitimate signal. More generally, it is possible to maximize the

secrecy rate through spatial beamforming. Even in challenging

environments with multiple eavesdroppers, spatial beamform-

ing can facilitate access security if there are enough spatial

degrees of freedom at the transmitter [201]. Unfortunately, the

secrecy performance of spatial beamforming heavily depends

on the accuracy of the CSI available at the multiple-antenna

transmitter. In general, the CSI of the eavesdropping channel

is difficult to obtain, since eavesdroppers usually hide their

existence by remaining silent (i.e., passive eavesdroppers).

In this case, artificial noise may be sent in the null space

of the legitimate devices’ channel matrices to confuse the

eavesdroppers [202].

In B5G wireless networks, the BTSs might be equipped with

a large-scale antenna array. By exploiting the very high spatial

resolution of the large-scale antenna array, secure access for a

massive number of devices can be provided. It has been proved

that if the BTS has full CSI, linear precoding can ensure that

the information leakage asymptotically tends to zero [203].

Hence, even without the eavesdroppers’ CSI, it is possible to

realize secure massive access. However, the acquisition of the

legitimate devices’ CSI in massive MIMO systems with a large

number of devices is not trivial [204]. Firstly, pilot sequences

are usually non-orthogonal, resulting in low CSI accuracy.

Secondly, the eavesdroppers can send interfering signals dur-

ing channel estimation to increase the interception probability.

Hence, providing physical-layer security in massive access is

still a challenging issue.

VIII. FUTURE RESEARCH DIRECTIONS

Despite the significant research efforts dedicated to fa-

cilitating massive access in B5G wireless networks, many

challenging issues remain to be tackled. In the following, we

discuss some future research directions.

A. Mobile Access

In cellular IoT, a fraction of the devices is expected to be

mobile and some devices may move with high speed. Mobility

gives rise to additional challenges for massive access. First,

mobility leads to fast time-varying channel fading making the

acquisition of the accurate CSI needed to facilitate massive

access very challenging, resulting in a performance degrada-

tion. Second, mobility causes frequent handoffs. For example,

in Internet-of-Vehicles applications, frequent handoffs between

BTSs may occur. Hence, the priority of mobility handoffs and

new access requests have to be properly handled. Moreover,

mobility may change the channel capacity of massive access

[205]. So far, only a few works have considered mobility in

massive access [206], [207].

B. Modulation and Coding

Modulation and coding schemes (MCS) are key for guaran-

teeing both high efficiency and high reliability for massive

access. The 5G wireless standard utilizes low-density par-

ity check (LDPC) codes and polar codes for the data and

control channels, respectively [16]. For massive access in

B5G wireless networks, some new characteristics have to be

considered. Firstly, the sporadic nature of IoT traffic favors

the use of short packets, and thus, short FEC codes should be

adopted. Secondly, as IoT devices are typically simple nodes

with limited computational capabilities, MCS in B5G wireless

networks have to be low-complexity. Therefore, the design of

new low-complexity short codes for massive access is a key

research problem.

C. Big Data Analytics and Large Dimensional Signal Process-

ing

In B5G wireless networks, there is a massive number of

IoT devices generating a huge volume of data. Meanwhile,

since the BTS is usually equipped with a large-scale antenna

array, the dimension of the received signal is very large.

Hence, massive access inevitably leads to big data in volume

and dimension. This significantly increases the burden on

B5G wireless networks. In order to improve the efficiency

of massive access, it is necessary to develop methods for big

data analytics and large dimensional signal processing. For

instance, a dimension reduction-based algorithm was designed

to decrease the computational complexity of massive active

device detection in B5G wireless networks [208]. However,

there is still a lack of efficient methods for channel estimation,

precoding design, and other aspects of massive access. Devel-

oping such methods is an exciting future research direction.

D. Ultra-Reliable Low-Latency Communication

Ultra-reliable low-latency communication (URLLC) is a

basic requirement in many cellular IoT application scenarios,

e.g. Internet-of-Vehicles [209]. However, it is very challenging

to guarantee URLLC over fading channels. First, massive

access leads to severe co-channel interference, which decreases

the access reliability. Second, short packets are used in cellular

IoT to decrease the latency, but they are also prone to a high

17

decoding error rate. Hence, achieving URLLC for massive

access is still an open issue.

E. Machine Learning-Based Massive Access

Smart communication is a new trend in wireless communi-

cations. Currently, machine learning, especially deep learning,

is being applied in wireless communications for resource allo-

cation, signal processing, channel estimation, and transceiver

design [210]. It has been shown that machine learning can

decrease the design complexity of wireless communication

networks while achieving high performance. In cellular IoT,

the BTS has to cope with the wireless access of a massive

number of devices, resulting in a high computational complex-

ity. The application of machine learning for massive access is

expected to significantly decrease complexity. Yet, there is a

lack of analytical frameworks for machine learning as applied

in wireless networks, which currently limits its applicability

in practice [211].

F. Convergence of Sensing, Computation, and Communication

Sensing, computation, and communication are three basic

functionalities of B5G wireless networks [212]. Traditionally,

these three functionalities have been carried out independently.

Hence, it is necessary to allocate wireless resources for each

functionality, resulting in a high resource consumption. In

the case of massive access, the resources required to support

these functionalities might be prohibitive. Hence, it is desirable

to jointly design these three functionalities to improve the

efficiency of massive access. For instance, the transmission

of sensed signals over multiple access channels can also

be exploited to perform computations by applying over-the-

air computation techniques in [213]. In this scenario, the

limited wireless resources can be utilized with high efficiency,

especially for massive access. Therefore, the convergence of

sensing, computation, and communication is an important

future direction for cellular IoT in B5G wireless networks.

IX. CONCLUSION

This paper provided a comprehensive review of massive

access in B5G wireless networks from different perspectives.

First, we summarized the basic characteristics of massive

access, such as low power, massive connectivity, and broad

coverage. Then, we surveyed information theoretical concepts

for massive access, focusing on massive random access and

massive short-packet access. Next, we discussed massive ac-

cess protocol design, with an emphasis on grant-free random

access protocols. In particular, we presented the sensing ma-

trix design and the corresponding device activity detection

algorithms, including optimization algorithms, greedy algo-

rithms, and Bayesian algorithms. Subsequently, we provided

an overview of massive orthogonal and non-orthogonal access

techniques, respectively. Furthermore, we identified challenges

for massive coverage enhancement in outdoor, indoor, and

rural environments. Finally, we discussed potential challenges

for providing massive access in B5G wireless networks and

pointed out some possible future research directions.

REFERENCES

[1] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi, “Internetof Things for smart cities,” IEEE Internet of Things J., vol. 1, no. 1, pp.22-32, Feb. 2014.

[2] N. Ahmed, D. De, and I. Hussain, “Internet of Things (IoT) for smartprecision agriculture and farming in rural areas,” IEEE Internet of ThingsJ., vol. 5, no. 6, pp. 4890-4899, Dec. 2018.

[3] L. D. Xu, W. He, and S. Li, “Internet of Things in industries: A survey,”IEEE Trans. Industiral Informatics, vol. 10, no. 4, pp. 2233-2243, Nov.2014.

[4] S. M. R. Islam, D. Kwak, M. H. Kabir, M. Hossain, and K-S. Kwak,“The Internet of Things for health care: A comprehensive survey,” IEEE

Access, vol. 3, pp. 678-708, Jun. 2015.

[5] Cisco, Cisco visual networking index: Global mobile data traffic forecastupdate, 2017-2022 white paper, Feb. 2019.

[6] Statista Research Department, “Internet of Things (IoT) connecteddevices installed base worldwide from 2015 to 2025 (in bil-lions),” [Online]: https://www.statista.com/statistics/471264/iot-number-of-connected-devices-worldwide, Nov. 2016.

[7] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and M. Ayyash,“Internet of Things: A survey on enabling technologies, protocols, andapplications,” IEEE Commun. Surv. Tuts., vol. 17, no. 4, pp. 2347-2376,Jun. 2015.

[8] S. G. Varghese, C. P. Kurian, V. I. George, A. John, V. Nayak, andA. Upadhyay, “Comparative study of zigbee topologies for IoT-basedlighting automation,” IET Wireless Sensor Syst., vol. 9, no. 4, pp. 201-207, Jul. 2019.

[9] K-H. Chang, “Bluetooth: A viable solution for IoT?” IEEE Wireless

Commun., vol. 21, no. 6, pp. 6-7, Dec. 2014.

[10] S. R. Pokhrel and C. Williamson, “Modeling compound TCP over WiFifor IoT,” IEEE/ACM Trans. Netw., vol. 26, no. 2, pp. 864-878, Apr. 2018.

[11] U. Raza, P. Kulkarni, and M. Sooriyabandara, “Low power wide areanetworks: An overview,” IEEE Commun. Surv. Tuts., vol. 19, no. 2, pp.855-873, Jun. 2017.

[12] J. P. S. Sundaram, W. Du, and Z. Zhao, “A survey on LoRa networking:Research problems, current solutions and open issues,” IEEE Commun.Surv. Tuts., vol. 22, no. 1, pp. 371-388, Mar. 2020.

[13] A. I. Sulyman, S. M. A. Oteafy, and H. S. Hassanein, “Expandingthe cellular-IoT umbrella: An architectural approach,” IEEE Wireless

Commun., vol. 24, no. 3, pp. 66-71, Jun. 2017.

[14] 3GPP TR 45.820, Technical specification group GSM/EDGE radioaccess network; Cellular system support for ultra-low complexity andlow throughput Internet of Things (CIoT), Nov. 2015.

[15] X. Chen, Massive Access for Cellular Internet of Things Theory andTechnique, Germany: Springer, 2019.

[16] Y. Cai, Z. Qin, F. Cui, G. Y. Li, and J. A. McCann, “Modulation andmultiple access for 5G networks,” IEEE Comm. Surv. Tuts., vol. 20, no.1, pp. 829-846, Mar. 2018.

[17] Y. Watanabe and K. Kamoi, “A formulation of the channel capacity ofmultiple-access channel,” IEEE Trans. Inf. Theory, vol. 55, no. 5, pp.2083-2096, May 2009.

[18] H. H. Permuter, T. Weissman, and J. Chen, “Capacity region of the finite-state multiple-access channel with and without feedback,” IEEE Trans.

Inf. Theory, vol. 55, no. 6, pp. 2455-2477, Jun. 2009.

[19] Y. Polyanskiy, H. V. Poor, and S. Verdu, “Channel coding rate in thefinite blocklength regime,” IEEE Trans. Inf. Theory, vol. 56, no. 5, Apr.2010.

[20] M. Hasan, E. Hossain, and D. Niyato, “Random access for machine-to-machine communication in LTE-advanced networks: Issues and ap-proaches,” IEEE Commun. Mag., vol. 51, no. 6, pp. 86-93, Jun. 2013.

[21] 3GPP TS 36.321 V9.3.0, “Evolved Universal Terrestrial Radio Access(E-UTRA); Medium Access Control (MAC),” Jun. 2010.

[22] N. H. Nguyen, B. Berscheid, and H. H. Nguyen, “Fast-OFDM withindex modulation for NB-IoT,” IEEE Commun. Lett., vol. 23, no. 7, pp.1157-1160, Jul. 2019.

[23] M. Zeng, A. Yadav, O. A. Dobre, G. I. Tsiropoulos, and H. V. Poor,“On the sum rate of MIMO-NOMA and MIMO-OMA systems,” IEEEWireless Commun. Lett., vol. 6, no. 4, pp. 534-537, Aug. 2017.

[24] J. Xu, J. Yao, L. Wang, Z. Ming, K. Wu, and L. Chen, “NarrowbandInternet of Things: Evolutions, technologies, and open issues,” IEEEInternet of Things J., vol. 5, no. 3, pp. 1149-1162, Jun. 2018.

[25] H-M. Wang, Q. Yang, Z. Ding, and H. V. Poor, “Secure short-packetcommunications for mission-critical IoT applications,” IEEE Trans. Wire-

less Commun., vol. 18, no. 5, pp. 2565-2578, May 2019.

18

[26] X. Chen and Y. Zhang, “Mode selection in MU-MIMO downlinknetworks: A physical layer security perspective,” IEEE Syst. J., vol. 11,no. 2, pp. 1128-1136, Jun. 2017.

[27] P. Popovski, J. J. Nielsen, C. Stefanovic, E. de Carvalho, E. Strom, K.F. Trillngsgaard, A-S. Bana, D. M. Kim, R. Kotaba, J. Park, and R. B.Sorensen, “Wireless access for ultra-reliable low-latency communication:Principles and building blocks,” IEEE Netw., vol. 32, no. 2, pp. 16-23,Mar. 2018.

[28] C. Bockelmann, N. Pratas, H. Nikopour, K. Au, T. Svensson, C.Stefanovic, P. Popovski, and A. Dekorsy, “Massive machine-type com-munications in 5G: Physical and MAC-layer solutions,” IEEE Commun.

Mag., vol. 54, no. 9, pp. 59-65, Sep. 2016.[29] Y. D. Beyene, R. Jantti, K. Ruttik, and S. Iraji, “On the performance

of narrow-band Internet of Things (NB-IoT),” in Proc. IEEE Wireless

Commun. Netw. Conf. (WCNC), San Francisco, CA, USA, 2017, pp. 1-6.[30] L. Liu, E. G. Larsson, W. Yu, P. Popovski, C. Stefanovic, and E. de

Carvalho, “Sparse signal processing for grant-free massive connectivity:A future paradigm for random access protocols in the Internet of Things,”IEEE Signal Process. Mag., vol. 35, no. 5, pp. 88-99, Sep. 2018.

[31] A-S. Bana, E. de Carvalho, B. Soret, T. Abrao, J. C. Marinello, E. G.Larsson, and P. Popovski, “Massive MIMO for Internet of Things (IoT)connectivity,” Physical Commun., vol. 35, pp. 1-17, Sep. 2019.

[32] Y. Liu, Z. Qin, M. Elkashlan, Z. Ding, A. Nallanathan, and L. Hanzo,“Nonorthogonal multiple access for 5G and beyond,” Proc. IEEE, vol.105, no. 12, pp. 2347-2381, Dec. 2017.

[33] Q. Qi, X. Chen, L. Lei, C. Zhong, Z. Zhang, “Outage-constrained robustdesign for sustainable B5G cellular Internet of Things,” IEEE Trans.Wireless Commun., vol. 18, no. 12, pp. 5780-5790, Dec. 2019.

[34] S. A. Busari, K. M. S. Huq, S. Mumtaz, and J. Rodriguez, “Terahertzmassive MIMO for beyond-5G wireless communications,” in Proc. IEEE

Intern. Conf. Commun. (ICC), Shanghai, China, Jun. 2019, pp. 1-6.[35] M. E. Morocho-Cayamcela, H. Lee, and W. Lim, “Machine learning for

5G/B5G mobile and wireless communications: Potential, limitations, andfuture directions,” IEEE Access, vol. 7, pp. 137184-137206, 2019.

[36] L. Zhang, Y-C. Liang, and D. Niyato, “6G visions: Mobile ultra-broadband, super Internet of Things, and artificial intelligent,” China

Commun., vol. 16, no. 8, pp. 1-14, Aug. 2019.[37] M. R. Palattella, M. Dohler, A. Grieco, G. Rizzo, J. Torsner, T. Engel,

and L. Ladid, “Internet of Things in the 5G era: Enablers, architecture,and business models,” IEEE J. Sel. Areas Commun., vol. 34, no. 3, pp.510-527, Mar. 2016.

[38] G. Durisi, T. Koch, and P. Popovski, “Toward massive, ultrareliable, andlow-latency wireless communication with short packets,” Proc. IEEE, vol.104, no. 9, pp. 1711-1726, Sep. 2016.

[39] L. Cheng, B. Henty, F. Bai, and D. D. Stancil, “Doppler spread andcoherence time of rural and highway vehicle-to-vehicle channels at5.9 GHz,” in Proc. IEEE Global Commun. Conf. (GLOBECOM), NewOrleans, LO, USA, Dec. 2008, pp. 1-6.

[40] Z. Tang, J. Wang, J. Wang, and J. Song, “Device activity detectionand non-coherent information transmission for massive machine-typecommunications,” IEEE Access, vol. 8, pp. 41452-41465, Feb. 2020.

[41] C. Zhu, J. Mietzner, and R. Schober, “On the performance of non-coherent transmission schemes with equal-gain combining in generalizedK-fading,” IEEE Trans. Wireless Commun., vol. 9, no. 4, pp. 1337-1349,Apr. 2010.

[42] S-C. Chang and E. Weldon, “Coding for T-user multiple-access chan-nels,” IEEE Trans. Inf. Theory, vol. 25, no. 6, pp. 684-691, Nov. 1979.

[43] J. A. Gubner, “On the capacity region of the discrete additive multiple-access arbitrarily varying channel,” IEEE Trans. Inf. Theory, vol. 38, no.4, pp. 1344-1347, Jul. 1992.

[44] J. Harshan and B. S. Rajan, “On two-user Gaussian multiple accesschannels with finite input constellations,” IEEE Trans. Inf. Theory, vol.57, no. 3, pp. 1299-1327, Mar. 2011.

[45] T. M. Cover and J. A. Thomas, Elements of Information Theory, 2nded. Hoboken, NJ, USA: Wiley, 2006.

[46] X. Chen and D. Guo, “Gaussian many-access channels: Definition andsymmetric capacity,” in Proc. IEEE Inf. Theory Workshop, Seville, Spain,Sep. 2013, pp. 1-5.

[47] X. Chen and D. Guo, “Many-access channels: The Gaussian case withrandom user activities,” in Proc. IEEE Int. Symp. Inf. Theory, Honolulu,HI, USA, Jun. 2014, pp. 3127-3131.

[48] X. Chen, T-Y. Chen, and D. Guo, “Capacity of Gaussian many-accesschannels,” IEEE Trans. Inf. Theory, vol. 63, no. 6, pp. 3516-3539, Jun.2017.

[49] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski,“Five disruptive technology directions for 5G,” IEEE Commun. Mag., vol.52, no. 2, pp. 74-80, Feb. 2014.

[50] V. W. S. Wong, R. Schober, D. W. K. Ng, and L.-C. Wang, Key

Technologies for 5G Wireless Systems. Cambridge, U.K.: CambridgeUniv. Press, 2017.

[51] F. Wei, Y. Wu, W. Chen, W. Yang, and G. Caire, “On the fundamentallimits of MIMO massive multiple access channels,” in Proc. IEEE Intern.

Conf. Commun. (ICC), Shanghai, China, May 2019, pp. 1-6.

[52] T. Kim and I. Bang, “An enhanced random access with preamble-assisted short-packet transmission for cellular IoT communications,”IEEE Commun. Lett., vol. 23, no. 6, pp. 1081-1084, Jun. 2019.

[53] L. Zhang and Y-C. Liang, “Average throughput analysis and optimizationin cooperative IoT networks with short packet communication,” IEEETrans. Veh. Technol., vol. 67, no. 12, pp. 11549-11562, Dec. 2018.

[54] G. Durisi, T. Koch, J. Ostman, Y. Polyanskiy, and W. Yang, “Short-packet communications over multiple-antenna Rayleigh-fading channels,”IEEE Trans. Commun., vol. 64, no. 2, pp. 618-629, Feb. 2016.

[55] S. Verdu, “Fifty years of Shannon theory,” IEEE Trans. Inf. Theory, vol.44, no. 6, pp. 2057-2078, Oct. 1998.

[56] L. Zheng and D. Tse, “Diversity and multiplexing: A fundamentaltradeoff in multiple-antenna channels,” IEEE Trans. Inf. Theory, vol. 49,no. 5, pp. 1073-1096, May 2003.

[57] G. Ozcan and M. C. Gursoy, “Throughput of cognitive radio systemswith finite blocklength codes,” IEEE J. Sel. Areas Commun., vol. 31, no.11, pp. 2541-2554, Nov. 2013.

[58] A. Rajandekar and B. Sikdar, “A survey of MAC layer issues andprotocols for machine-to-machine communications,” IEEE Internet of

Things J., vol. 2, no. 2, pp. 175-186, Apr. 2015.

[59] C. Tang, L. Song, J. Bagadeesh, S. Wu, S. Biaz, Q. Yang, and H. Wang,“Comparative investigation on CSMA/CA-based opportunistic randomaccess for Internet of Things,” IEEE Internet of Things J., vol. 1, no.2, pp. 171-179, Apr. 2019.

[60] H. Seo, J-P. Hong, and W. Choi, “Low latency random access forsporadic MTC devices in Internet of Things,” IEEE Internet of Things J.,vol. 6, no. 3, pp. 5108-5118, Jun. 2019.

[61] Y. Polyanskiy, “A perspective on massive random access,” in Proc. IEEE

Intern. Symp. Inf. Theory (ISIT), Aachen, Germany, Jun. 2017, pp. 2523-2527.

[62] M. Centenaro, L. Vangelista, S. Saur, A. Weber, and V. Braun, “Compar-ison of collision-free and contention-based radio access protocols for theInternet of Things,” IEEE Trans. Commun., vol. 65, no. 9, pp. 3832-3846,Sep. 2017.

[63] A. T. Abebe and C. G. Kang, “Comprehensive grant-free random accessfor massive & low latency communication,” in Proc. IEEE Intern. Conf.

Commun. (ICC), Paris, France, May 2017, pp. 1-6.

[64] X. Shao, X. Chen, C. Zhong, J. Zhao, and Z. Zhang, “Protocol designand analysis for cellular Internet of Things with massive access,” in Proc.IEEE Intern. Conf. Commun. (ICC), Shanghai, China, May 2019, pp. 1-6.

[65] L. Liu and W. Yu, “Massive device connectivity with massive MIMO,”in Proc. IEEE Intern. Symp. Inf. Theory (ISIT), Aachen, Germany, Jun.2017, pp. 1-5.

[66] K. Senel and E. G. Larsson, “Grant-free massive MTC-enabled massiveMIMO: A compressive sensing approach,” IEEE Trans. Commun., vol.66, no. 12, pp. 6164-6175, Dec. 2018.

[67] Z. Chen, F. Sohrabi, and W. Yu, “Sparse activity detection for massiveconnectivity,” IEEE Trans. Signal Process., vol. 66, no. 7, pp. 1890-1904,Apr. 2018.

[68] J. Ding and J. Choi, “Comparison of preamble structures for grant-freerandom access in massive MIMO systems,” IEEE Wireless Commun. Lett.,vol. 9, no. 2, pp. 166-170, Feb. 2020.

[69] L. Liu and W. Yu, “Massive connectivity with massive MIMO-part I:Device activity detection and channel estimation,” IEEE Trans. Signal

Process., vol. 66, no. 11, pp. 2933-2946, Jun. 2018.

[70] X. Shao, X. Chen, C. Zhong, J. Zhao, and Z. Zhang, “A unified designof massive access for cellular Internet of Things,” IEEE Internet of ThingsJ., vol. 6. no. 2, pp. 3934-3947, Apr. 2019.

[71] J. Wang, Z. Zhang, and L. Hanzo, “Joint active user detection andchannel estimation in massive access systems exploiting Reed-Mullersequences,” IEEE J. Sel. Topics Signal Process., vol. 13, no. 3, pp. 739-752, Jun. 2019.

[72] S. Li, W. Zhang, Y. Cui, H. V. Cheng, and W. Yu, “Joint design ofmeasurement matrix and sparse support recovery method via deep auto-encoder”, IEEE Signal Process. Lett., vol. 26, no. 12, pp. 1778-1782,Dec. 2019.

[73] S. Mousavi, M. M. R. Taghiabadi, and R. Ayanzadeh, “A survey oncompressive sensing: Classical results and recent advancements,” Aug.2019. [Online]: arXiv:1908.01014v1.

19

[74] D. L. Donoho and J. Tanner, “Sparse nonnegative solution of underde-termined linear equations by linear programming,” Proc. Nat. Acad. Sci.USA, vol. 102, no. 27, pp. 9446-9451, Jul. 2005.

[75] E. J. Candes, J. K. Romberg, and T. Tao, “Stable signal recovery fromincomplete and inaccurate measurements,” Commun. Pure Appl. Math.,vol. 59, no. 8, pp. 1207-1223, Aug. 2006.

[76] S. J. Kim, K. Koh, M. Lustig, S. Boyd, and D. Gorinevsky, “An interior-point method for large-scale l1-regularized least squares,” IEEE J. Sel.

Topics Signal Process., vol. 1, no. 4, pp. 606-617, Dec. 2007.

[77] M. Golbabaee and P. Vandergheynst, “Compressed sensing of simulta-neous low-rank and joint-sparse matrices,” [Online]: arXiv preprint arXiv:1211.5058, Nov. 2012.

[78] X. Shao, X. Chen, and R. Jia, “Low-complexity design of massive devicedetection via Riemannian pursuit,” in Proc. IEEE Global Commun. Conf.

(GLOBECOM), Waikoa, USA, Dec. 2019, pp. 1-6.

[79] C. Bockelmann, H. F. Schepker, and A. Dekorsy, “Compressive sens-ing based multi-user detection for machine-to-machine communication,”Trans. Emerging Telecommun. Technol., vol. 24, no. 4, pp. 389-400, Apr.2013.

[80] I. Roth, M. Kliesch, G. Wunder, and J. Eisert. “Reliable recovery ofhierarchically sparse signals and application in machine-type communi-cations,” [Online]: arXiv preprint arXiv:1612.07806, Dec. 2016.

[81] G. Wunder, I. Roth, R. Fritschek, and J. Eisert “Performance ofhierarchical sparse detectors for massive MTC,” [Online]: arXiv preprintarXiv:1806.02754, Oct. 2018.

[82] X. Xu, X. Rao, and V. K. N. Lau, “Active user detection and channelestimation in uplink CRAN systems,” Proc. IEEE Inter. Conf. Commun.

(ICC), London, UK, pp. 2727-2732, Jun. 2015.

[83] Z. Utkovski, O. Simeone, T. Dimitrova, and P. Popovski, “Random ac-cess in C-RAN for user activity detection with limited-capacity fronthaul,”IEEE Signal Process. Lett., vol. 24, no. 1, pp. 17-21, Jan. 2017.

[84] X. Shao, X. Chen, C. Zhong, and Z. Zhang, “Joint activity detection andchannel estimation for mmW/THz wideband massive access,” in Proc.

IEEE Intern. Conf. Commun. (ICC), Dublin, Ireland, Jun. 2020, pp. 1-6.

[85] M. Ke, Z. Gao, Y. Wu, X. Gao, and R. Schober, “Compressive sensing-based adaptive active user detection and channel estimation: Massiveaccess meets massive MIMO,” IEEE Trans. Signal Process., vol. 68, pp.764-779, Jan. 2020.

[86] Q. He, T. Q. S. Quek, Z. Chen, Q. Zhang, and S. Li, “Compressive chan-nel estimation and multi-user detection in C-RAN with low complexitymethods,” IEEE Trans. Wireless Commun., vol. 17, no. 6, pp. 3931-3944,Jun. 2018.

[87] L. Liu and W. Yu, “Massive connectivity with massive MIMO-part II:Achievable rate characterization,” IEEE Trans. Signal Process., vol. 66,no. 11, pp. 2947-2959, Jun. 2018.

[88] J. Ahn, B. Shim, and K. B. Lee, “EP-based joint active user detectionand channel estimation for massive machine-type communications,” IEEETrans. Commun., vol. 67, no. 7, pp. 5178-5189, Jul. 2019.

[89] Z. Sun, Z. Wei, L. Yang, J. Yuan, X. Cheng, and L. Wan, “Exploitingtransmission control for joint user identification and channel estimationin massive connectivity,” IEEE Trans. Commun., vol. 67, no. 9, pp. 6311-6326, Sep. 2019.

[90] J. Zhang, Y. Pan, and J. Xu, “Compressive sensing for joint user activityand data detection in grant-free NOMA,” IEEE Wireless Commun. Lett.,vol. 8, no. 3, pp. 857-860, Jun. 2019.

[91] B. K. Jeong, B. Shim, and K. B. Lee, “MAP-based active user and datadetection for massive machine-type communications,” IEEE Trans. Veh.

Technol., vol. 67, no. 9, pp. 8481-8494, Sep. 2018.

[92] B. Wang, L. Dai, Y. Zhang, T. Mir, and J. Li, “Dynamic compressivesensing-based multi-user detection for uplink grant-free NOMA,” IEEECommun. Lett., vol. 20, no. 11, pp. 2320-2323, Nov. 2016.

[93] Y. Du, B. Dong, Z. Chen, X. Wang, Z. Liu, P. Gao, and S. Li, “Efficientmulti-user detection for uplink grant-free NOMA: Prior-information aidedadaptive compressive sensing perspective,” IEEE J. Sel. Areas Commun.,vol. 35, no. 12, pp. 2812-2828, Dec. 2017.

[94] S. Haghighatshoar, P. Jung, and G. Caire, “Improved scaling law foractivity detection in massive MIMO systems,” in Proc. IEEE Intern.

Symp. Inf. Theory (ISIT), Vail, USA, Jun. 2018, pp. 381-385.

[95] A. Fengler, S. Haghighatshoar, P. Jung, and G. Caire, “Non-bayesianactivity detection, large-scale fading coefficient estimation, and un-sourced random access with a massive MIMO receiver,” [Online]:http://arxiv.org/abs/1910.11266, Oct. 2019.

[96] Z. Chen, F. Sohrabi, Y. Liu, and W. Yu, “Covariance based joint activityand data detection for massive random access with massive MIMO,” inProc. IEEE Intern. Conf. Commun. (ICC), Shanghai, China, 2019, pp.1-6.

[97] A. Fengler, G. Caire, P. Jung, and S. Haghighatshoar, “Massive MIMOunsourced random access,” [Online]: http://arxiv.org/abs/1901.00828, Jan.2019.

[98] V. K. Amalladinne, J-F. Chamberland, and K. R. Narayanan, “A codedcompresssed sensing scheme for uncoordinated multiple access,” [On-line]: https://arxiv.org/abs/1809.04745, Sep. 2018.

[99] E. de Carvalho, E. Bjornson, J. H. Sørensen, E. G. Larsson, and P.Popovski, “Random pilot and data access in massive MIMO for machine-type communications,” IEEE Trans. Wireless Commun., vol. 16, no. 12,pp. 7703-7717, Dec. 2017.

[100] Y. Han, B. D. Rao, and J. Lee, “Massive uncoordinated access withmassive MIMO: A dictionary learning approach,” IEEE Trans. Wireless

Commun., vol. 19, no. 2, pp. 1320-1332, Feb. 2020.

[101] T. L. Marzetta, “Noncooperative cellular wireless with unlimited num-bers of base station antennas,” IEEE Trans. Wireless Commun., vol. 9,no. 11, pp. 3590-3600, Nov. 2010.

[102] H. Q. Ngo, E. Larsson, and T. Marzetta, “Energy and spectral efficiencyof very large multiuser MIMO systems,” IEEE Trans. Commun., vol. 61,no. 4, pp. 1436-1449, Apr. 2013.

[103] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resourceallocation in OFDMA systems with large numbers of base stationantennas,” IEEE Trans. Wireless Commun., vol. 11, no. 9, pp. 3292-3304,Sep. 2012.

[104] R. Deng, Z. Jiang, S. Zhou, and Z. Niu, “How often should be CSI beupdated for massive MIMO systems with massive connectivity?” in Proc.

IEEE Global Commun. Conf. (GLOBECOM), Singapore, Singapore, Dec.2017, pp. 1-6.

[105] D. J. Love, R. W. Heath, V. K. N. Lau, D. Gesbert, B. D. Rao, and M.Andrews, “An overview of limited feedback in wireless communicationsystems,” IEEE J. Sel. Areas Commun., vol. 26, no. 8, pp. 1341-1365,Oct. 2008.

[106] J-C. Shen, J. Zhang, E. Alsusa, and K. B. Letaief, “Compressed CSIacquisition in FDD massive MIMO: How much training is needed?” IEEE

Trans. Wireless Commun., vol. 15, no. 6, pp. 4145-4156, Jun. 2016.

[107] Z. Gao, L. Dai, W. Dai, B. Shim, and Z. Wang, “Structured compressivesensing-based spatio-temporal joint channel estimation for FDD massiveMIMO,” IEEE Trans. Commun., vol. 64, no. 2, pp. 601-617, Feb. 2016.

[108] W. Shen, L. Dai, B. Shim, S. Mumtaz, and Z. Wang, “Joint CSITacquisition based on low-rank matrix completion for FDD massive MIMOsystems,” IEEE Commun. Lett., vol. 19, no. 12, pp. 2178-2181, Dec. 2015.

[109] C-K. Wen, W-T. Shih, and S. Jin, “Deep learning for massive MIMOCSI feedback,” IEEE Wireless Commun. Lett., vol. 7, no. 5, pp. 748-751,Oct. 2018.

[110] T. L. Marzetta, E. G. Larsson, H. Yang, and H. Q. Ngo, Fundamental

of Massive MIMO, UK: Cambridge University Press, 2016.

[111] O. Elijah, C. Y. Leow, T. A. Rahman, S. Nunoo, and S. Z. Iliya,“A comprehensive survey of pilot contamination in massive MIMO-5Gsystem,” IEEE Commun. Surv. Tuts., vol. 18, no. 2, pp. 905-923, Jun.2016.

[112] H. V. Cheng, E. Bjornson, and E. G. Larsson, “Optimal pilot andpayload power control in single-cell massive MIMO systems,” IEEE

Trans. Signal Process., vol. 65, no. 9, pp. 2363-2378, May 2017.

[113] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resourceallocation in multi-cell OFDMA systems with limited backhaul capacity,”IEEE Trans. Wireless Commun., vol. 11, no. 10, pp. 3618–3631, Sep.2012.

[114] L. Liang, W. Xu, and X. Dong, “Low-complexity hybrid precoding inmassive multiuser MIMO systems,” IEEE Wireless Commun. Lett., vol.3, no. 6, pp. 653-656, Dec. 2014.

[115] F. Sohrabi and W. Yu, “Hybrid digital and analog beamforming designfor large-scale antenna arrays”, IEEE J. Sel. Topics Signal Process., vol.10, no. 3, pp. 501-513, April 2016.

[116] Z. Wei, D. W. K. Ng, and J. Yuan, “NOMA for hybrid mmwavecommunication systems with beamwidth control,” IEEE J. Sel. Topics

in Signal Process., vol. 13, no. 3, pp. 567–583, Mar. 2019.

[117] Q. Shi and M. Hong, “Spectral efficiency optimization for millimeterwave multiuser MIMO systems,” IEEE J. Sel. Topics Signal Process., vol.12, no. 3, pp. 455-468, Jun. 2018.

[118] A. Liu, V. K. N. Lau, and M-J. Zhao, “Stochastic successive convexoptimization for two-timescale hybrid precoding in massive MIMO,”IEEE J. Sel. Topics Signal Process., vol. 12, no. 3, pp. 432-444, Jun.2018.

[119] L. Fan, S. Jin, C-K. Wen, and H. Zhang, “Uplink achievable rate formassive MIMO systems with low-resolution ADC,” IEEE Commun. Lett.,vol. 19, no. 12, pp. 2186-2189, Dec. 2015.

20

[120] S. Wang, Y. Li, and J. Wang, “Multiuser detection in massive spatialmodulation MIMO with low-resolution ADCs,” IEEE Trans. WirelessCommun., vol. 14, no. 4, pp. 2156-2168, Apr. 2015.

[121] G. Yu, X. Chen, and D. W. K. Ng, “Low-cost design of massive accessfor cellular Internet of Things,” IEEE Trans. Commun., vol. 67, no. 11,pp. 8008-8020, Nov. 2019.

[122] W. Roh, J-Y. Seol, J. Park, B. Lee, J. Lee, Y. Kim, J. Cho, K.Cheun, and F. Aryanfar, “Millimeter-wave beamforming as an enablingtechnology for 5G cellular communications: Theoretical feasibility andprototype results,” IEEE Commun. Mag., vol. 52, no. 2, pp. 106-113,Feb. 2014.

[123] R. W. Heath, N. Gonzalez-Prelcic, S. Rangan, W. Roh, and A. M.Sayeed, “An overview of signal processing techniques for millimeter waveMIMO systems,” IEEE J. Sel. Topics Signal Process., vol. 10, no. 3, pp.436-452, Apr. 2016.

[124] T. S. Rappaport, Y. Xing, G. R. MacCartney, A. F. Molisch, E. Mellios,and J. Zhang, “Overview of millimeter wave communications for fifth-generation (5G) wireles networks-with a focus on propagation model,”IEEE Trans. Ant. Propa., vol. 65, no. 12, pp. 6213-6230, Dec. 2017.

[125] C. Han, Y. Wu, Z. Chen, and X. Wang, “Terahertz communications(TeraCom): Challenges and impact on 6G wireless systems,” [Online]:arXiv:1912.06040v2, Dec. 2019.

[126] Z. Zhang, Y. Liang, W. Shi, L. Yuan, and G. Gui, “l1,2-regularization-based super-resolution sparse channel estimation for mmwave massiveMIMO systems,” IEEE Access, vol. 7, pp. 75837-75844, 2019.

[127] F. Bellili, F. Sohrabi, and W. Yu, “Generalized approximate messagepassing for massive MIMO mmwave channel estimation with Laplacianprior,” IEEE Trans. Commun., vol. 67, no. 5, pp. 3205-3219, May 2019.

[128] X. Hu, C. Zhong, X. Chen, W. Xu, and Z. Zhang, “Cluster groupingand power control for angle-domain mmwave MIMO NOMA systems,”IEEE J. Sel. Topics Signal Process., vol. 13, no. 5, pp. 1167-1180, Sep.2019.

[129] C. Liu, M. Li, S. V. Hanly, I. B. Collings, and P. Whiting, “Millimeterwave beam alignment: Large deviations analysis and design insights,”IEEE J. Sel. Areas Commun., vol. 35, no. 7, pp. 1619-1631, Jul. 2017.

[130] L. Dai, B. Wang, Y. Yuan, S. Han, C. L. I, and Z. Wang, “Non-orthogonal multiple access for 5G: Solutions, challenges, opportunities,and future research trends,” IEEE Commun. Mag., vol. 53, no. 9, pp.74-81, Sep. 2015.

[131] M. Shirvnimoghaddam, M. Condoluci, M. Dohler, and S. J. Johnson,“On the fundamental limits of random non-orthogonal multiple access incellular massive IoT,” IEEE J. Sel. Areas Commun., vol. 35, no. 10, pp.2238-2252, Oct. 2017.

[132] Z. Ding, X. Lei, G. K. Karagiannidis, R. Schober, J. Yuan, and V. K.Bhargava, “A survey on non-orthogonal multiple access for 5G networks:Research challenges and future trends,” IEEE J. Sel. Areas Commun., vol.35, no. 10, pp. 2181-2195, Oct. 2017.

[133] Z. Wei, D. W. K. Ng, J. Yuan, and H. Wang, “Optimal resourceallocation for power-efficient MC-NOMA with imperfect channel stateinformation,” IEEE Trans. Commun., vol. 65, no. 9, pp. 3944–3961, Sep.2017.

[134] X. Chen, Z. Zhang, C. Zhong, and R. Jia, “On the design of mas-sive access,” in Proc. Intern. Conf. Wireless Commun. Signal Process.

(WCSP), Nanjing, China, Oct. 2017, pp. 1-6.

[135] S. M. R. Islam, N. Avazov, O. A. Dobre, and K-S. Kwak, “Power-domain non-orthogonal multiple access (NOMA) in 5G: potentials andchallenges,” IEEE Commun. Surv. Tuts., vol. 19, no. 2, pp. 721-742, Jun.2017.

[136] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance ofnonorthogonal multiple access in 5G systems with randomly deployedusers,” IEEE Signal Process. Lett., vol. 21, no. 12, pp. 1501-1505, Dec.2014.

[137] J. Choi, “Power allocation for max-sum rate and max-min rate pro-portional fairness in NOMA,” IEEE Commun. Lett., vol. 20, no. 10, pp.2055-2058, Oct. 2016.

[138] X. Chen and R. Jia, “Exploiting rateless coding for massive access,”IEEE Trans. Veh. Technol., vol. 67, no. 11, pp. 11253-11257, Nov. 2018.

[139] Y. Liu, M. Elkashlan, Z. Ding, and G. K. Karagiannidis, “Fairness ofuser clustering in MIMO non-orthogonal multiple access systems,” IEEECommun. Lett., vol. 20, no. 7, pp. 1465-1468, Jul. 2016.

[140] Y. Sun, D. W. K. Ng, J. Zhu, and R. Schober, “Robust and secureresource allocation for full-duplex MISO multicarrier NOMA aystems,”IEEE Trans. Commun., vol. 66, no. 9, pp. 4119-4137, Sep. 2018.

[141] X. Chen, Z. Zhang, C. Zhong, and D. W. K. Ng, “Exploiting multiple-antenna for non-orthogonal multiple access,” IEEE J. Sel. Areas Com-

mun., vol. 35, no. 10, pp. 2207-2220, Oct. 2017.

[142] F. Tian and X. Chen, “Multiple-antenna techniques in nonorthogonalmultiple access: A review,” Front. Inform. Technol. Electron. Eng., vol.20, no. 12, pp. 1665-1697, Dec. 2019.

[143] X. Chen, Z. Zhang, C. Zhong, R. Jia, and D. W. K. Ng, “Fully non-orthogonal communication for massive access,” IEEE Trans. Commun.,vol. 66, no. 4, pp. 1717-1731, Apr. 2018.

[144] X. Chen, R. Jia, and D. W. K. Ng, “On the design of massivenon-orthogonal multiple access with imperfect successive interferencecancellation,” IEEE Trans. Commun., vol. 67, no. 3, pp. 2539-2551, Mar.2019.

[145] R. Jia, X. Chen, C. Zhong, D. W. K. Ng, H. Lin, and Z. Zhang, “Designof non-orthogonal beamspace multiple access for cellular Internet-of-Things,” IEEE J. Sel. Topics Signal Process., vol. 13, no. 3, pp. 538-552,Jun. 2019.

[146] L. Dai, B. Wang, Z. Ding, Z. Wang, S. Chen, and L. Hanzo, “A surveyof non-orthogonal multiple access for 5G,” IEEE Commun. Surv. Tuts.,vol. 20, no. 3, pp. 2294-2323, Sep. 2018.

[147] Y. Du, B. Dong, P. Gao, Z. Chen, J. Fang, and S. Wang, “Low-complexity LDS-CDMA detection based on dynamic factor graph,” inProc. IEEE Global Commun. Conf. Workshops (GC Wkshps), Washington,USA, Dec. 2016, pp. 1-6.

[148] R. Razavi, M. AL-Imari, M. A. Imaran, R. Hoshyar, and D. Chen, “Onreceiver design for uplink low density signature OFDM (LDS-OFDM),”IEEE Trans. Commun., vol. 60, no. 11, pp. 3499-3508, Nov. 2012.

[149] Z. Yuan, C. Yan, Y. Yuan, and W. Li, “Blind multiple user detection forgrant-free MUSA without reference signal,” in Proc. IEEE Veh. Technol.Conf. (VTC), Toronto, Canada, Sep. 2017, pp. 1-5.

[150] M. Vameghestahbanati, I. D. Marsland, R. H. Gohary, and H.Yanikomeroglu, “Multimensional constellations for uplink SCMAsystems-a comparative study,” IEEE Commun. Surv. Tuts., vol. 21, no.3, pp. 2169-2194, Sep. 2019.

[151] F. Wei and W. Chen, “Low complexity iterative receiver design forsparse code multiple access,” IEEE Trans. Commun., vol. 65, no. 2, pp.621-634, Feb. 2017.

[152] F. Wei, W. Chen, Y. Wu, J. Ma, and T. A. Tsiftsis, “Message-passingreceiver design for joint channel estimation and data decoding in uplinkgrant-free SCMA systems,” IEEE Trans. Wireless Commun., vol. 18, no.1, pp. 167-181, Jan. 2019.

[153] E. Becirovic, E. Bjornson, and E. G. Larsson, “How much will tinyIoT nodes profit from massive base station arrays,” in Proc. European

Signal Processing Conference (EUSIPCO), Rome, Italy, Sep. 2018, pp.832-836.

[154] J. Zhang, S. Chen, Y. Lin, J. Zheng, B. Ai, and L. Hanzo, “Cell-freemassive MIMO: A new next-generation paradigm,” IEEE Access, vol. 7,pp. 99878-99888, Jul. 2019.

[155] P. Liu, K. Luo, D. Chen, and T. Jiang, “Spectral efficiency analysisof cell-free massive MIMO systems with zero-forcing detector,” IEEE

Trans. Wireless Commun., vol. 19, no. 2, pp. 795-807, Feb. 2020.

[156] H. Q. Ngo, A. Ashikhmin, H. Yang, E. Larsson, and T. L. Marzetta,“Cell-free massive MIMO versus small cells,” IEEE Trans. Wireless

Commun., vol. 16, no. 3, pp. 1834-1850, Mar. 2017.

[157] T. Nakamura, S. Nagata, A. Benjebbour, Y. Kishiyama, T. Hai, X.Shen, N. Yang, and N. Li, “Trends in small cell enhancements in LTEadvanced,” IEEE Commun. Mag., vol. 51, no. 2, pp. 98-105, Feb. 2013.

[158] E. Nayebi, A. Ashikhmin, T. L. Marzetta, H. Yang, and B. D. Rao,“Precoding and power optimization in cell-free massive MIMO systems,”IEEE Trans. Wireless Commun., vol. 16, no. 7, pp. 4445-4459, Jul. 2017.

[159] X. Hu, C. Zhong, X. Chen, W. Xu, H. Lin, and Z. Zhang, “Cell-free massive MIMO systems with low resolution ADCs,” IEEE Trans.

Commun., vol. 67, no. 10, pp. 6844-6857, Oct. 2019.

[160] S. Feng, P. Setoodeh, and S. Haykin, “Smart home: Cognitive inter-active people-centric Internet of Things,” IEEE Commun. Mag., vol. 55,no. 2, pp. 34-39, Feb. 2017.

[161] L. Subrt and P. Pechac, “Intelligent walls as autonomous parts of smartindoor environments,” IET Commun., vol. 6, no. 8, pp. 1004-1010, Aug.2012

[162] X. Tan, Z. Sun, J. M. Jornet, and D. Pados, “Increasing indoor spectrumsharing capacity using smart reflect-array,” in Proc. IEEE Intern. Conf.

Commun. (ICC), Kuula Lumpur, May 2016, pp. 1-6.

[163] D. Xu, X. Yu, Y. Sun, D. W. K. Ng, and R. Schober, “Resourceallocation for secure IRS-assisted multiuser MISO systems,” in Proc.IEEE Global Commun. (GLOBECOM), Waikoloa, HI, USA, Dec. 2019,pp. 1-6.

[164] X. Yu, D. Xu, Y. Sun, D. W. K. Ng, and R. Schober, “Robust andsecure wireless communications via intelligent reflecting surface,” IEEE

J. Sel. Areas Commun., vol. PP, no. 99, pp. 1-1, 2020.

21

[165] C. Huang, A. Zappone, M. Debbah, and C. Yuen, “Achievable ratemaximization by passive intelligent mirrors,” in Proc. IEEE Intern. Conf.Acous. Speech Signal Process. (ICCASP), Calgary, Canada, Apr. 2018,pp. 3714-3718.

[166] Q. Wu and R. Zhang, “Beamforming optimization for intelligentreflecting surface with discrete phase shifts,” in Proc. IEEE Intern. Conf.

Acous. Speech Signal Process. (ICCASP), Brighton, UK, May 2019, pp.1-7.

[167] E. Basar, M. D. Renzo, J. D. Rosny, M. Debbah, M-S. Alouini, andR. Zhang, “Wireless communications through reconfigurable intelligentsurface,” IEEE Access, vol. 7, pp. 116753-116773, Aug. 2019.

[168] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wirelessnetwork via joint active and passive beamforming,” IEEE Trans. Wireless

Commun., vol. 18, no. 11, pp. 5394-5409, Nov. 2019.

[169] G. Yu, X. Chen, C. Zhong, D. W. K. Ng, and Z. Zhang, “Design,analysis and optimization of a large intelligent reflecting surface aidedB5G cellular Internet of Things,” IEEE Internet of Things J., vol. PP, no.99, pp. 1-1, 2020.

[170] N. Ahmed, D. De, and L. Hussain, “Internet of Things (IoT) for smartprecision agriculture and farming in rural areas,” IEEE Internet of ThingsJ., vol. 5, no. 6, pp. 4890-4899, Dec. 2018.

[171] H. Kaur and S. K. Sood, “Energy-efficient IoT-fog-cloud architecturalparadigm for real-time wildfire prediction and forecasting,” IEEE Syst.

J., vol. 14, no. 2, pp. 2003-2011, Jun. 2020.

[172] T. Qiu, Z. Zhao, T. Zhang, C. Chen, and C. L. P. Chen, “UnderwaterInternet of Things in smart ocean: System architecture and open issues,”IEEE Trans. Indus. Inf., vol. 16, no. 7, pp. 4297-4307, Jul. 2020.

[173] R. Briskman, “Domestic satellite services for rural areas,” IEEE Com-mun. Mag., vol. 22, no. 3, pp. 35-38, Mar. 1984.

[174] O. Kodheli, A. Guidotti, A. Vanelli-Coralli, “Integration of satellite in5G through LEO constellations,” in Proc. IEEE Global Commun. Conf.

(GLOBECOM), Sigapore, Singapore, Dec. 2017, 1-6.

[175] B. Di, L. Song, Y. Li, and H. V. Poor, “Ultra-dense LEO: integration ofsatellite access networks into 5G and beyond,” IEEE Wireless Commun.,vol. 26, no. 2, pp. 62-69, Apr. 2019.

[176] O. Kodheli, S. Andrenacci, N. Maturo, S. Chatzinotas, and F. Zimmer,“An uplink UE ground-based scheduling techniques for 5G mMTCsystems over LEO satellite,” IEEE Access, vol. 7, pp. 67413-67427, May2019.

[177] G. Zheng, S. Chatzinotas, and B. Ottersten, “Generic optimization oflinear precoding in multibeam satellite systems”, IEEE Trans. WirelessCommun., vol. 11, no. 6, pp. 2308-2320, Jun. 2012.

[178] L. You, A. Liu, W. Wang, and X. Gao, “Outage constrained robustmultigroup multicast beamforming for multi-beam satellite communica-tion systems,” IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 352-355,Apr. 2019.

[179] X. Zhu, C. Jiang, L. Yin, L. Kuang, N. Ge, and J. Lu, “Cooperativemultigroup multicast transmission in integrated terrestrial-satellite net-works,” IEEE J. Sel. Areas Commun., vol. 36, no. 5, pp.981-992, May2018.

[180] M. D. Sanctis, E. Cianca, G. Araniti, I. Bisio, and R. Prasad, “Satellitecommunications supporting internet of remote things,” IEEE Internet ofThings J., vol. 3, no. 1, pp. 113-123, Feb. 2016.

[181] X. Sun, N. Yang, S. Yan, Z. Ding, D. W. K. Ng, C. Shen, and Z.Zhong, “Joint beamforming and power allocation in downlink NOMAmultiuser MIMO networks,” IEEE Trans. Wireless Commun., vol. 17, no.8, pp. 5367-5381, Aug. 2018.

[182] M. Tian, Q. Zhang, S. Zhao, Q. Li, and J. Qin, “Robust beamformingin downlink MIMO NOMA networks using cutting-set method,” IEEE

Commun. Lett., vol. 22, no. 3, pp. 547-577, Mar. 2018.

[183] Y. Sun, D. W. K. Ng, J. Zhu, and R. Schober, “Robust and secureresource allocaton for full-duplex MISO multicarrier NOMA systems,”IEEE Trans. Commun., vol. 66, no. 9, pp. 4119-4137, Sep. 2018.

[184] X. Chen, Z. Zhang, H-H. Chen, and H. Zhang, “Enhancing wirelessinformation and power transfer by exploiting multi-antenna techniques,”IEEE Commun. Mag., vol. 53, no. 4, pp. 133-141, Apr. 2015.

[185] E. Boshkovska, D. W. K. Ng, N. Zlatanov, and R. Schober, “Practicalnon-linear energy harvesting model and resource allocation for SWIPTsystems”, IEEE Commun. Lett., vol. 19, no. 12, pp. 2082-2085, Dec.2015.

[186] C. Zhong, X. Chen, Z. Zhang, and G. Karagiannidis, “Wireless poweredcommunications: Performance analysis and optimization,” IEEE Trans.

Commun., vol. 63, no. 12, pp. 5178-5190, Dec. 2015.

[187] W. Na, J. Park, C. Lee, K. Park, J. Kim, and S. Cho, “Energy-efficient mobile charging for wireless power transfer in Internet of Thingsnetworks,” IEEE Internet of Things J., vol. 5, no. 1, pp. 79-92, Jan. 2018.

[188] D. Zhai, H. Chen, Z. Lin, Y. Li, and B. Vucetic, “Accumulatethen transmit: Multiuser scheduling in full-duplex wireless-powered IoTsystems,” IEEE Internet of Things J., vol. 5, no. 4, pp. 2753-2767, Aug.2018.

[189] X. Chen, C. Yuen, and Z. Zhang, “Wireless energy and informationtransfer tradeoff for limited feedback multi-antenna systems with energybeamforming,” IEEE Trans. Veh. Technol., vol. 63, no. 1, pp. 407-412,Jan. 2014.

[190] D. W. K. Ng, E. S. Lo, and R. Schober, “Robust beamforming forsecure communication in systems with wireless information and powertransfer,” IEEE Trans. Wireless Commun., vol. 13, no. 8, pp. 4599–4615,Aug. 2014.

[191] X. Chen, X. Wang, and X. Chen, “Energy-efficient optimization forwireless information and power transfer in large-scale MIMO systemsemploying energy beamforming,” IEEE Wireless Commun. Lett., vol. 2,no. 6, pp. 667-670, Dec. 2013.

[192] Z. Chen, P. Xu, Z. Ding, and X. Dai, “Cooperative transmission insimultaneous wireless inforamtion and power transfer networks,” IEEE

Trans. Veh. Technol., vol. 65, no. 10, pp. 8710-8715, Oct. 2016.

[193] H. Liang, C. Zhong, X. Chen, H. Suraweera, and Z. Zhang, “Wirelesspowered dual-hop multi-antenna relaying systems: Impact of CSI andantenna correlation,” IEEE Trans. Wireless Commun., vol. 16, no. 4, pp.2505-2519, Apr. 2017.

[194] Q. Qi and X. Chen, “Wireless powered massive access for cellularInternet of Things with imperfect SIC and non-linear EH,” IEEE Internet

of Things J., vol. 6, no. 2, pp. 3110-3120, Apr. 2019.

[195] W. Stallings, Cryptography and Network Security Principles and Prac-

tices, Prentice Hall PTR, 2006.

[196] J. L. Massey, “An introduction to contemporary cryptology,” Proc.

IEEE, vol. 76, no. 5, pp. 533-549, May 1988.

[197] Q. Qi, X. Chen, C. Zhang, and Z. Zhang, “Physical layer securityfor massive access in cellular Internet of Things,” SCIENCE CHINA Inf.

Sciences, vol. 63, no. 2, pp. 1-12, Feb. 2020.

[198] F. Shu, X. Wu, J. Hu, J. Li, R. Chen, and J. Wang, “Secure and precisewireless transmission for random-subcarrier-selection-based directionalmodulation,” IEEE J. Sel. Areas Commun., vol. 36, no. 4, pp. 890-904,Jul. 2018.

[199] Y. Cai, Z. Wei, R. Li, D. W. K. Ng, and J. Yuan, “Joint trajectory andresource allocation design for energy-efficient secure UAV communica-tion systems,” IEEE Trans. Commun., vol. PP, no. 99, pp. 1-1, 2020.

[200] X. Chen, D. W. K. Ng, W. Gerstacker, and H-H. Chen, “A survey ofmultiple-antenna techniques for physical layer security,” IEEE Commun.

Surv. Tuts., vol. 19, no. 2, pp. 1027-1053, Jun. 2017.

[201] W. Wang, K. C. Teh, and K. H. Li, “Secrecy throughput maximizationfor MISO multi-eavesdropper wiretap channels,” IEEE Trans. Inf. Foren.

Security, vol. 12, no. 3, pp. 505-515, Mar. 2017.

[202] X. Zhang, M. R. McKay, X. Zhou, and R. W. Heath, “Artificial-noise-aided secure multi-antenna transmission with limited feedback,” IEEE

Trans. Wireless Commun., vol. 14, no. 5, pp. 2742-2754, May 2015.

[203] X. Chen, C. Yuen, and Z. Zhang, “Exploiting large-scale MIMOtechniques for physical layer security with imperfect channel state infor-mation,” in Proc. IEEE Global Commun. Conf. (GLOBECOM), Austin,USA, Dec. 2014.

[204] X. Chen, Z. Zhang, C. Zhong, D. W. K. Ng, and R. Jia, “Exploit-ing inter-user interference for secure massive non-orthogonal multipleaccess,” IEEE J. Sel. Areas Commun., vol. 36, no. 4, pp. 788-801, Apr.2018.

[205] Z. Zhang, C. Jiao, and C. Zhong, “Impact of mobility on the uplinksum rate of MIMO-OFDMA cellular systems,” IEEE Trans. Commun.,vol. 65, no. 10, pp. 4218-4231, Oct. 2017.

[206] H. Zhang, H. Dai, Z. Zhang, and Y. Huang, “Mobile conductance insparse networks and mobility-connectivity tradeoff,” IEEE Trans. Wireless

Commun., vol. 15, no. 4, pp. 2954-2965, Apr. 2016.

[207] Z. Ding, R. Schober, P. Fan, and H. V. Poor, “OTFS-NOMA: Anefficient approach for exploiting heterogeneous user mobility profiles,”IEEE Trans. Commun., vol. 67, no. 11, pp. 7950-7965, Nov. 2019.

[208] X. Shao, X. Chen, and R. Jia, “A dimension reduction-based jointactivity detection and channel estimation algorithm for massive access,”IEEE Trans. Signal Process., vol. 68, pp. 420-435, 2020.

[209] M. Bennis, M. Debbah, and H. V. Poor, “Ultrareliable and low-latencywireless communication: Tail, risk, and scale,” Proc. IEEE, vol. 106, no.10, pp. 1834-1853, Oct. 2018.

[210] T. Wang, C-K. Wen, H. Wang, F. Gao, T. Jiang, and S. Jin, “Deeplearning for wireless physical layer: Opportunities and challenges,” China

Commun., vol. 14, no. 11, pp. 92-111, Nov. 2017.

22

[211] S-L. Huang, X. Xu, L. Zheng, and G. W. Wornell, “An informationtheoretic interpretation to deep neural networks,” in Proc. IEEE Int. Symp.Inf. Theory (ISIT), Paris, France, Jul. 2019, pp. 1984-1988.

[212] G. Zhu and K. Huang, “MIMO over-the-air computation for high-mobility multimodal sensing,” IEEE Internet of Things J., vol. 6, no.4, pp. 6089-6103, Aug. 2019.

[213] X. Chen and Q. Qi, Convergence of Energy, Communication and

Computation in B5G Cellular Internet of Things, Germany: Springer,2020.